{
 "site": "https://agentapilist.com",
 "verified": "2026-10-16",
 "count": 22,
 "apis": [
  {
   "slug": "openrouter-armor",
   "name": "OpenRouter Armor",
   "vendor": "OpenRouter",
   "category": "Guardrail / tool-call policy",
   "filter": "guardrail",
   "verified": "2026-10-16",
   "summary": "New (beta, launched 2026-10-15): an OpenRouter plugin that checks every tool call an agent’s model makes against your plain-English rules and blocks out-of-policy calls before they reach your app. OpenRouter says it uses TypeSafe’s Jev decision model. Free during the beta.",
   "url": "https://agentapilist.com/apis/openrouter-armor/",
   "cost_calculator": null,
   "priced": false,
   "not_comparable": "Free during the beta; OpenRouter says GA pricing will be announced later, and your normal model inference is billed separately.",
   "pricing": [
    {
     "text": "Armor evaluation is not billed during the beta. OpenRouter says pricing will be announced before general availability. Normal model inference is still billed.",
     "verification": "Verified 2026-10-16 on openrouter.ai/docs/guides/features/plugins/armor"
    }
   ],
   "limits": [
    {
     "text": "Added latency per tool call: 279 ms median and 334 ms p90 (OpenRouter’s own measurement with its Claude Code compatible rule set).",
     "verification": "Verified 2026-10-16 on openrouter.ai/docs/guides/features/plugins/armor"
    },
    {
     "text": "Rule fields: id up to 64 chars (unique), name up to 120 chars, instructions up to 8,000 chars, optional threshold 0–1.",
     "verification": "Verified 2026-10-16 on openrouter.ai/docs/guides/features/plugins/armor"
    },
    {
     "text": "Not yet supported for HIPAA, Zero Data Retention, or data-region-restricted requests. A plugin request in those contexts gets a 403. Claude Code requests are released with an unavailable verdict.",
     "verification": "Verified 2026-10-16 on openrouter.ai/docs/guides/features/plugins/armor"
    },
    {
     "text": "Coverage: any model OpenRouter serves (docs, for Claude Code auto mode); the launch post says any agent doing inference through OpenRouter can opt in. On Anthropic endpoints that support the beta, Anthropic’s own classifier verdict is passed through instead.",
     "verification": "Verified 2026-10-16 on openrouter.ai/docs/guides/features/plugins/armor"
    }
   ],
   "status": [
    {
     "text": "Beta since October 2026. The config fields, safeguard_results, and the openrouter:armor_blocked item type may still change.",
     "verification": "Verified 2026-10-16 on openrouter.ai/docs/guides/features/plugins/armor"
    },
    {
     "text": "Launched 2026-10-15. OpenRouter’s post says Armor “uses @typesafeai’s Jev model”. The Armor docs only say “the decision model” and do not name a model or version.",
     "verification": "Verified 2026-10-16 on OpenRouter’s official X posts (2026-10-15)"
    },
    {
     "text": "OpenRouter claims that on an internal test set it built, Jev 1.13 had the highest accuracy and lowest cost compared with other Decisions models and Claude Code’s use of Sonnet 5. No test set or metrics are published.",
     "verification": "Not published by vendor: no test set or metrics behind this claim (checked 2026-10-16 on https://openrouter.ai/docs/guides/features/plugins/armor)"
    }
   ],
   "auth": [
    {
     "text": "OpenRouter API key sent as Authorization: Bearer $OPENROUTER_API_KEY (Armor is a plugin on normal OpenRouter requests).",
     "verification": "Verified 2026-10-16 on openrouter.ai/docs/guides/features/plugins/armor"
    }
   ],
   "fit": [
    "Good for a drop-in “auto mode” for agents that already call models through OpenRouter, with no classifier to host. Run mode: \"audit\" on real traffic first, then switch to enforce.",
    "Expect some false positives and misses (OpenRouter says so). Keep the environment facts current, since rules that refer to them depend on them.",
    "With claude_code_compatibility on Messages, a client that ignores safeguard_results will still run flagged calls. If you don’t control the client, use output_mode \"text\".",
    "If you need your own thresholds, a human-review path, or traffic outside OpenRouter, call Jev directly. OpenRouter’s cookbook shows an approve, block, or review gate."
   ],
   "sources": [
    {
     "title": "OpenRouter docs: Armor",
     "url": "https://openrouter.ai/docs/guides/features/plugins/armor"
    },
    {
     "title": "OpenRouter launch post on X (2026-10-15)",
     "url": "https://x.com/openrouter/status/2110838205654470840"
    },
    {
     "title": "OpenRouter post: opt in with { id: \"armor\" }",
     "url": "https://x.com/OpenRouter/status/2110838279566479831"
    },
    {
     "title": "OpenRouter post: Jev 1.13 internal-test claim",
     "url": "https://x.com/OpenRouter/status/2110838266920591515"
    },
    {
     "title": "OpenRouter docs: Gate Agent Tool Calls with Jev",
     "url": "https://openrouter.ai/docs/cookbook/building-agents/gate-tool-calls-with-jev"
    }
   ]
  },
  {
   "slug": "openai-decisions",
   "name": "OpenAI Decisions API",
   "vendor": "OpenAI",
   "category": "Decision model",
   "filter": "decision",
   "verified": "2026-10-16",
   "summary": "Dedicated POST /v1/decisions endpoint that returns typed answers (probability, choice, or score) about text and images instead of generated text. Model: gpt-6-luna (only model available). Since Oct 15, 2026 GPT-Live can call it natively during voice sessions.",
   "url": "https://agentapilist.com/apis/openai-decisions/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/openai-decisions/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "$0.10 per 1M input tokens; no output, cache-read, or cache-write charges.",
     "verification": "Verified 2026-10-16 on OpenAI Decisions guide"
    },
    {
     "text": "Regional processing premiums and long-context input multipliers apply (amounts not stated on the guide). Other gpt-6-luna requests outside /v1/decisions use normal model pricing.",
     "verification": "Verified 2026-10-16 on OpenAI Decisions guide"
    }
   ],
   "limits": [
    {
     "text": "Context window: not stated on the Decisions guide. Rate limits are set per usage tier and shown in your organization’s limits page; no Decisions-specific public table.",
     "verification": "Not published by vendor (checked 2026-10-16 on https://developers.openai.com/api/docs/guides/decisions and https://developers.openai.com/api/docs/guides/rate-limits)"
    }
   ],
   "status": [
    {
     "text": "Public beta (released Oct 6, 2026 per the API changelog); OpenAI says it expects GA \"in the coming weeks\".",
     "verification": "Verified 2026-10-16 on OpenAI Decisions guide and API changelog"
    },
    {
     "text": "New Oct 15, 2026: native Decisions integration in GPT-Live (gpt-live-1). Live can delegate a request to Decisions directly, or your session can trigger decisions on input/output transcript deltas. Needs api.decisions.write permission.",
     "verification": "Verified 2026-10-16 on OpenAI API changelog and “Connect voice to Decisions” guide"
    },
    {
     "text": "Data controls: ZDR and HIPAA for eligible customers; data residency/regional processing in the US and Europe (EEA + Switzerland).",
     "verification": "Verified 2026-10-16 on OpenAI Decisions guide"
    },
    {
     "text": "No published SLA, dated model snapshot, or deprecation schedule for /v1/decisions.",
     "verification": "Not published by vendor (checked 2026-10-16 on https://developers.openai.com/api/docs/guides/decisions; enterprise contracts not visible to us)"
    }
   ],
   "auth": [],
   "fit": [
    "Good fit: classification, routing, prioritization, and review queues with app-owned thresholds set from your own labeled examples.",
    "Only the alias gpt-6-luna is documented — behavior can change under you. Run a canary eval and log the model string on every call.",
    "No rationale is returned; if users need an explanation, pair it with a generative model and human review.",
    "OpenAI Usage Policies restrict automating high-stakes decisions in sensitive areas (e.g. critical infrastructure, product safety components, medical) without human review. Keep it advisory; never the sole authority."
   ],
   "sources": [
    {
     "title": "OpenAI Decisions guide",
     "url": "https://developers.openai.com/api/docs/guides/decisions"
    },
    {
     "title": "OpenAI API changelog",
     "url": "https://developers.openai.com/api/docs/changelog"
    },
    {
     "title": "OpenAI guide: Connect voice to Decisions",
     "url": "https://developers.openai.com/api/docs/guides/decisions-voice"
    },
    {
     "title": "OpenAI community beta announcement (2026-10-06)",
     "url": "https://community.openai.com/t/decisions-api-is-now-available-in-public-beta/1403877"
    },
    {
     "title": "OpenAI Usage Policies",
     "url": "https://openai.com/policies/usage-policies"
    }
   ]
  },
  {
   "slug": "microsoft-decision-1",
   "name": "Microsoft-Decision-1",
   "vendor": "Microsoft (Foundry)",
   "category": "Decision model",
   "filter": "decision",
   "verified": "2026-10-16",
   "summary": "Small decision model in Microsoft Foundry that returns typed numeric decisions (noul / choice / score) about text or JSON state. No free-form output or rationale. Also callable through OpenRouter and Vercel AI Gateway without an Azure deployment.",
   "url": "https://agentapilist.com/apis/microsoft-decision-1/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/microsoft-decision-1/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "$0.042 per 1M input tokens; output $0 (OpenRouter listing, provider Azure).",
     "verification": "Verified 2026-10-16 on OpenRouter; Foundry/Azure price page not checked"
    }
   ],
   "limits": [
    {
     "text": "32,768-token context window (OpenRouter).",
     "verification": "Verified 2026-10-16 on OpenRouter"
    },
    {
     "text": "Rate limits are per deployment quota (429 on exceed); specific numbers not published on the how-to page.",
     "verification": "Verified 2026-10-16 on Microsoft Learn how-to"
    }
   ],
   "status": [
    {
     "text": "Released Oct 9, 2026 (OpenRouter). On 2026-10-16 OpenRouter showed 100.00% uptime and 99.99% availability over the prior 3 days, and a 0.20 s p50 latency (1-week, best provider).",
     "verification": "Verified 2026-10-16 on OpenRouter model page"
    },
    {
     "text": "OpenRouter: \"Weights are updated continually while the API shape stays the same.\" Microsoft docs list model version 1.",
     "verification": "Verified 2026-10-16; whether Foundry pins weights per version is unclear"
    },
    {
     "text": "Release stage: public preview, per Microsoft’s Foundry announcement. Neither the Learn how-to page nor the announcement publishes an SLA for this model.",
     "verification": "Verified 2026-10-16 on https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-microsoft-decision-1-in-microsoft-foundry-for-decision-and-classific/4562 (SLA not published by vendor on these pages)"
    }
   ],
   "auth": [],
   "fit": [
    "Good fit: gates, routing, ranking, grading, and agent/tool selection inside a fixed allowlist.",
    "Microsoft lists known risks: bias, wording/order sensitivity, calibration strongest on familiar tasks, no explanations, may miss harmful content.",
    "Microsoft: for consequential decisions about people (credit, employment, housing, healthcare, legal) use it for decision support with meaningful human review, not as the sole decision-maker.",
    "Recommended: include a \"cannot tell\" option, randomize option order, and set thresholds from labeled data."
   ],
   "sources": [
    {
     "title": "Microsoft Learn: Deploy and use Microsoft-Decision-1",
     "url": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision"
    },
    {
     "title": "OpenRouter model page",
     "url": "https://openrouter.ai/microsoft/microsoft-decision-1"
    },
    {
     "title": "Vercel changelog: Microsoft Decision-1 on AI Gateway (2026-10-09)",
     "url": "https://vercel.com/changelog/microsoft-decision-1-now-available-on-ai-gateway"
    }
   ]
  },
  {
   "slug": "vercel-ai-gateway-decisions",
   "name": "Vercel AI Gateway — Decisions endpoint",
   "vendor": "Vercel",
   "category": "Gateway / proxy",
   "filter": "gateway",
   "verified": "2026-10-16",
   "summary": "OpenAI-compatible POST /decisions endpoint on Vercel AI Gateway that routes to any decision model on the gateway (e.g. openai/gpt-6-luna-decisions, typesafe-ai/jev, microsoft/microsoft-decision-1).",
   "url": "https://agentapilist.com/apis/vercel-ai-gateway-decisions/",
   "cost_calculator": null,
   "priced": false,
   "not_comparable": "Pass-through: no markup, you pay the routed provider’s list price, so the cost depends on which model you route to.",
   "pricing": [
    {
     "text": "No markup and no platform fee on tokens: you pay the provider’s list price (also with your own key). Per-request cost is reported in response metadata; the docs’ example shows $0.10 per 1M for openai/gpt-6-luna-decisions, matching OpenAI’s list price.",
     "verification": "Verified 2026-10-16 on https://vercel.com/docs/ai-gateway/pricing and https://vercel.com/docs/ai-gateway/sdks-and-apis/openai-decisions"
    },
    {
     "text": "A request with refused questions is billed for its input like any other request.",
     "verification": "Verified 2026-10-16 on Vercel Decisions docs"
    }
   ],
   "limits": [
    {
     "text": "choice: 1–255 choices; score: 2–10 levels; up to 128 images per request (base64 data URLs only).",
     "verification": "Verified 2026-10-16 on Vercel Decisions docs"
    },
    {
     "text": "Text-only models such as typesafe-ai/jev return 400 on image requests.",
     "verification": "Verified 2026-10-16 on Vercel Decisions docs"
    }
   ],
   "status": [
    {
     "text": "Documentation page live; release stage not stated on the page. OpenAI Decisions added Oct 7, 2026 and Microsoft Decision-1 added Oct 9, 2026 (Vercel changelog).",
     "verification": "Verified 2026-10-16 on Vercel docs and changelog"
    }
   ],
   "auth": [],
   "fit": [
    "Useful for vendor-diverse fallbacks behind one API shape — use fallbacks for availability, not as a vote that overrides your rules.",
    "Adds a third party in the data path; confirm ZDR and provider restrictions match your requirements."
   ],
   "sources": [
    {
     "title": "Vercel docs: OpenAI-compatible Decisions API",
     "url": "https://vercel.com/docs/ai-gateway/sdks-and-apis/openai-decisions"
    },
    {
     "title": "Vercel changelog: OpenAI Decisions API on AI Gateway (2026-10-07)",
     "url": "https://vercel.com/changelog/openai-decisions-api-now-available-on-ai-gateway"
    },
    {
     "title": "Vercel changelog: Microsoft Decision-1 on AI Gateway (2026-10-09)",
     "url": "https://vercel.com/changelog/microsoft-decision-1-now-available-on-ai-gateway"
    }
   ]
  },
  {
   "slug": "jev-typesafe-ai",
   "name": "Jev (TypeSafe AI)",
   "vendor": "TypeSafe AI",
   "category": "Decision model",
   "filter": "decision",
   "verified": "2026-10-16",
   "summary": "Hosted \"System One\" decision model from TypeSafe AI, and the model OpenRouter says powers its new OpenRouter Armor tool-call guard (beta, 2026-10-15). Send a state plus typed questions (noul, choice, score) to POST /v1/systemone and get probabilities back. jev-latest points to the stable jev-1.13.0; jev-1.14.0 is the preview (jev-preview).",
   "url": "https://agentapilist.com/apis/jev-typesafe-ai/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/jev-typesafe-ai/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "$0.042 per 1M input tokens ($42 per billion) for both Jev 1.14 and 1.13; output tokens free. Long-context mode costs the same per token.",
     "verification": "Verified 2026-10-16 on docs.typesafe.ai/models (no change)"
    }
   ],
   "limits": [
    {
     "text": "64k tokens per request / 32k per question by default; 262k for both with the opt-in long-context header (slower and less accurate per TypeSafe).",
     "verification": "Verified 2026-10-16 on docs.typesafe.ai/models (no change)"
    },
    {
     "text": "English is the primary training language and where accuracy is best; other languages (including CJK) are handled less well, per TypeSafe.",
     "verification": "Verified 2026-10-16 on docs.typesafe.ai/models"
    }
   ],
   "status": [
    {
     "text": "Correction: jev-latest resolves to jev-1.13.0, the most recent stable release. jev-1.14.0 (available from October 14, 2026) is preview only, via the jev-preview alias.",
     "verification": "Verified 2026-10-16 on docs.typesafe.ai/models (no change)"
    },
    {
     "text": "Not trained on customer requests; ZDR for enterprise. No SLA stated on the models page.",
     "verification": "Verified 2026-10-15 on docs.typesafe.ai/models"
    },
    {
     "text": "Now powers OpenRouter Armor (beta, launched 2026-10-15), per OpenRouter’s launch post. Armor checks agent tool calls against plain-English rules and is free during the beta. See the OpenRouter Armor page linked below.",
     "verification": "Verified 2026-10-16 on OpenRouter’s official X posts (2026-10-15)"
    },
    {
     "text": "Also on OpenRouter: typesafe/jev-1.14 and typesafe/jev-1.13, billed to your OpenRouter account, via POST /api/alpha/decisions or /api/v1/systemone. On OpenRouter, ~typesafe/jev-latest tracks the newest release (1.14); TypeSafe’s own jev-latest points to 1.13.0.",
     "verification": "Verified 2026-10-16 on openrouter.ai/docs/guides/community/jev"
    },
    {
     "text": "Correction: an earlier version of this page showed $0.42/1M taken from jevtypesafeai.com, a domain we could not tie to TypeSafe. TypeSafe’s own docs show $0.042.",
     "verification": "Correction logged 2026-10-15"
    }
   ],
   "auth": [],
   "fit": [
    "Pin a versioned ID (e.g. jev-1.13.0) rather than jev-latest or jev-preview so thresholds don’t drift with version changes.",
    "Text-oriented; check image support for your version in TypeSafe docs before relying on it."
   ],
   "sources": [
    {
     "title": "TypeSafe docs: Models",
     "url": "https://docs.typesafe.ai/models"
    },
    {
     "title": "OpenRouter docs: Jev",
     "url": "https://openrouter.ai/docs/guides/community/jev"
    },
    {
     "title": "OpenRouter docs: Armor",
     "url": "https://openrouter.ai/docs/guides/features/plugins/armor"
    },
    {
     "title": "OpenRouter Armor launch post (names Jev)",
     "url": "https://x.com/openrouter/status/2110838205654470840"
    }
   ]
  },
  {
   "slug": "cloudflare-clef",
   "name": "Cloudflare Clef (Clef, Clef-flash, Clef-omni)",
   "vendor": "Cloudflare (Workers AI)",
   "category": "Decision model",
   "filter": "decision",
   "verified": "2026-10-16",
   "summary": "Cloudflare-hosted decision models on Workers AI that take a state and a schema of typed questions (noul, choice, score) and return a probability for every allowed option. Three variants: Clef (27B), Clef-flash (9B), Clef-omni (30B MoE, 3B active).",
   "url": "https://agentapilist.com/apis/cloudflare-clef/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/cloudflare-clef/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "Clef $0.240, Clef-omni $0.150, Clef-flash $0.038 per 1M input tokens; no output-token charge. Media are converted to input tokens.",
     "verification": "Verified 2026-10-16 on Cloudflare Workers AI pricing page (no change; page last updated Oct 15, 2026)"
    },
    {
     "text": "Workers AI billing: $0.011 per 1,000 Neurons; 10,000 Neurons/day free on Free and Paid plans (limits reset 00:00 UTC). Clef-flash is listed at 3,455 Neurons per 1M input tokens.",
     "verification": "Verified 2026-10-16 on Cloudflare Workers AI pricing page"
    }
   ],
   "limits": [
    {
     "text": "Context: Clef 65,536 tokens; Clef-flash 24,576; Clef-omni 64,000.",
     "verification": "Verified 2026-10-15 on Cloudflare model pages"
    },
    {
     "text": "Clef-omni: 1–64 questions; up to 4 images (4 MiB each), 4 audio clips, 2 videos (60 s each). Each image capped at 1,024 tokens.",
     "verification": "Verified 2026-10-15 on clef-omni model page"
    }
   ],
   "status": [
    {
     "text": "Listed on Workers AI model catalog and pricing page.",
     "verification": "Verified 2026-10-15 against official page"
    }
   ],
   "auth": [
    {
     "text": "Cloudflare API token as Authorization: Bearer, on the account-scoped REST endpoint /accounts/{ACCOUNT_ID}/ai/run/@cf/cloudflare/clef; or a Workers AI binding inside a Worker.",
     "verification": "Verified 2026-10-16 on developers.cloudflare.com/workers-ai/models/clef"
    }
   ],
   "fit": [
    "Clef-flash is the cheapest Clef variant but has the smallest context; long text state is truncated to fit. (Correction 2026-10-16: an earlier version called it the cheapest hosted option here; Perplexity, Inception, and Liquid d1 list $0.02 per 1M.)",
    "Test calibration and order sensitivity on your own labels; vendor pages publish no accuracy figure."
   ],
   "sources": [
    {
     "title": "Workers AI pricing",
     "url": "https://developers.cloudflare.com/workers-ai/platform/pricing/"
    },
    {
     "title": "clef model page",
     "url": "https://developers.cloudflare.com/workers-ai/models/clef/"
    },
    {
     "title": "clef-flash model page",
     "url": "https://developers.cloudflare.com/workers-ai/models/clef-flash/"
    },
    {
     "title": "clef-omni model page",
     "url": "https://developers.cloudflare.com/workers-ai/models/clef-omni/"
    }
   ]
  },
  {
   "slug": "perplexity-decisions",
   "name": "Perplexity Decisions API (pplx-decider)",
   "vendor": "Perplexity",
   "category": "Decision model",
   "filter": "decision",
   "verified": "2026-10-17",
   "summary": "Perplexity API endpoint (POST https://api.perplexity.ai/v1/decisions) that answers yes/no, multiple-choice, and scored questions about text, JSON, and images with probabilities. $0.02 per 1M input tokens on pplx-decider-v1.2 ($0.01 on the 4B Flash model), output free.",
   "url": "https://agentapilist.com/apis/perplexity-decisions/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/perplexity-decisions/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "$0.02 per 1M input tokens on pplx-decider-v1.2 and $0.01 on pplx-decider-flash-v1.2; output tokens free; no per-request fee. State, questions, and images count as input. In Perplexity’s tests an image cost about 1,000 input tokens per megapixel.",
     "verification": "Verified 2026-10-17 on https://docs.perplexity.ai/getting-started/pricing and https://docs.perplexity.ai/docs/decisions/quickstart"
    }
   ],
   "limits": [
    {
     "text": "1–128 questions per request; 1–255 options per choice; 1–10 score levels; 256K-token context window (input must be under 256K tokens per request); up to 8 images per question; 32 MiB body; images up to 2,048 tiles of 32×32 px (e.g. 1440×1440); oversize images time out with 504 after about a minute.",
     "verification": "Verified 2026-10-16 on Perplexity Decisions quickstart (full page)"
    },
    {
     "text": "Rate limit: 10 requests per second per organization on every plan, plus a token limit for large bursts; 429 responses include Retry-After.",
     "verification": "Verified 2026-10-16 on Perplexity Decisions quickstart"
    },
    {
     "text": "Correction: on pplx-decider-v1.2 and flash-v1.2 the shared state is read and billed once per request, so grouping questions costs less than separate requests. An earlier version of this page, based on a community forum post, said state was billed once per question.",
     "verification": "Verified 2026-10-16 on https://docs.perplexity.ai/docs/decisions/quickstart"
    }
   ],
   "status": [
    {
     "text": "pplx-decider-v1.2 replaced pplx-decider-v1.1-27b and pplx-decider-flash-v1.2 is new, per the October 2026 changelog entry (no exact day shown). Release-stage label and SLA not stated.",
     "verification": "Verified 2026-10-17 on https://docs.perplexity.ai/docs/resources/changelog and https://docs.perplexity.ai/docs/decisions/quickstart"
    },
    {
     "text": "Hugging Face model cards exist for pplx-decider-v1.2 (Qwen3.8-27B backbone) and pplx-decider-flash-v1.2 (Qwen3.5-4B), but both cards say the repository is private and needs an access token. The weights are not openly downloadable, so this is not an open-weight model.",
     "verification": "Verified 2026-10-17 on https://huggingface.co/perplexity-ai/pplx-decider-v1.2 and https://huggingface.co/perplexity-ai/pplx-decider-flash-v1.2"
    },
    {
     "text": "Benchmarks: the model cards report Perplexity’s own “Decision Index 0.3” scores (v1.2 68.09, Flash 58.88, vendor-run). Artificial Analysis’s Decision Index lists only the older Decider v1.1 27B (53.1); v1.2 is not on it yet.",
     "verification": "Verified 2026-10-17 on https://huggingface.co/perplexity-ai/pplx-decider-v1.2 and https://artificialanalysis.ai/evaluations/artificial-analysis-decision-index"
    }
   ],
   "auth": [],
   "fit": [
    "Lowest listed per-token price among hosted decision APIs here; validate accuracy on your own labels.",
    "The 10 req/s org-wide cap is low for high-volume pipelines; batch several questions into one request where they share state."
   ],
   "sources": [
    {
     "title": "Perplexity API pricing",
     "url": "https://docs.perplexity.ai/getting-started/pricing"
    },
    {
     "title": "Perplexity changelog",
     "url": "https://docs.perplexity.ai/docs/resources/changelog"
    },
    {
     "title": "Perplexity Decisions quickstart",
     "url": "https://docs.perplexity.ai/docs/decisions/quickstart"
    },
    {
     "title": "Hugging Face: pplx-decider-v1.2 (private repo)",
     "url": "https://huggingface.co/perplexity-ai/pplx-decider-v1.2"
    },
    {
     "title": "Hugging Face: pplx-decider-flash-v1.2 (private repo)",
     "url": "https://huggingface.co/perplexity-ai/pplx-decider-flash-v1.2"
    },
    {
     "title": "Artificial Analysis Decision Index",
     "url": "https://artificialanalysis.ai/evaluations/artificial-analysis-decision-index"
    },
    {
     "title": "Community forum thread (unofficial)",
     "url": "https://community.perplexity.ai/t/decisions-api-shared-state-counted-per-question-in-usage-input-tokens/6316"
    }
   ]
  },
  {
   "slug": "inception-mercury-decide",
   "name": "Inception Mercury Decide",
   "vendor": "Inception",
   "category": "Decision model",
   "filter": "decision",
   "verified": "2026-10-16",
   "summary": "Decision model served at POST /v1/decisions (model mercury-decide) that evaluates named questions against a shared state and returns typed answers with probabilities.",
   "url": "https://agentapilist.com/apis/inception-mercury-decide/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/inception-mercury-decide/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "Input listed as $0.04 → $0.02 per 1M tokens (shown as a 50%-off promotion); cached input and output $0.",
     "verification": "Verified 2026-10-15 on Inception models page; promotion end date not stated"
    }
   ],
   "limits": [
    {
     "text": "choice: 1–255 options; score: 1–10 levels; up to 16 images per request, 10 MiB each (base64 PNG/JPEG/WebP/GIF, no remote URLs); 20 MiB body limit (413 if over). No chat, streaming, or tools.",
     "verification": "Verified 2026-10-15 on Inception Decisions guide"
    },
    {
     "text": "Context window: not published (listed as “—” on the models page).",
     "verification": "Not published by vendor (checked 2026-10-16 on https://docs.inceptionlabs.ai/get-started/models)"
    }
   ],
   "status": [
    {
     "text": "Listed on Inception models page with a \"50% off Mercury Decide\" banner.",
     "verification": "Verified 2026-10-15 against official page"
    }
   ],
   "auth": [
    {
     "text": "Inception API key as Authorization: Bearer $INCEPTION_API_KEY on POST https://api.inceptionlabs.ai/v1/decisions.",
     "verification": "Verified 2026-10-16 on docs.inceptionlabs.ai/capabilities/decisions"
    }
   ],
   "fit": [
    "Promotional price may revert; budget at the undiscounted rate."
   ],
   "sources": [
    {
     "title": "Inception models",
     "url": "https://docs.inceptionlabs.ai/get-started/models"
    },
    {
     "title": "Inception Decisions guide",
     "url": "https://docs.inceptionlabs.ai/capabilities/decisions"
    }
   ]
  },
  {
   "slug": "liquid-ai-d1",
   "name": "Liquid AI d1",
   "vendor": "Liquid AI",
   "category": "Decision model",
   "filter": "decision",
   "verified": "2026-10-16",
   "summary": "Hosted decision model on the Liquid API (System One shape) returning noul, choice, and score answers with zero generated tokens; supports images. Open-weight variants also documented.",
   "url": "https://agentapilist.com/apis/liquid-ai-d1/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/liquid-ai-d1/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "$0.02 per 1M input tokens, $0 output, on Liquid’s pricing page. New orgs get $2 free credit on Liquid Console.",
     "verification": "Verified 2026-10-15 on liquid.ai/pricing"
    },
    {
     "text": "Correction: an earlier version of this page reported a price conflict ($0.04 per 1M on Liquid’s d1 blog and Vercel’s model page vs $0.02 on Liquid’s pricing page). On 2026-10-16 Liquid’s pricing page lists $0.02, the d1 blog states no per-token price, and Vercel’s public models API lists $0.02 per 1M input for liquid/d1.",
     "verification": "Verified 2026-10-16 on https://www.liquid.ai/pricing, https://www.liquid.ai/blog/d1-decision-model and https://ai-gateway.vercel.sh/v1/models"
    }
   ],
   "limits": [
    {
     "text": "Up to 8 images; body under 4.5 MB; at most 10,000 32×32 patches; image tokens are charged again for every question. d1:free is text-only.",
     "verification": "Verified 2026-10-15 on Liquid docs: d1"
    },
    {
     "text": "Context window: Liquid’s docs do not publish one; Vercel (third party) lists 65,536 in its models API.",
     "verification": "Not published by vendor (checked 2026-10-16 on https://docs.liquid.ai/llms.txt); third-party figure confirmed on https://ai-gateway.vercel.sh/v1/models"
    }
   ],
   "status": [
    {
     "text": "Available on the Liquid API; Liquid’s blog says d1 is also on Vercel and OpenRouter (text only there).",
     "verification": "Verified 2026-10-15 on Liquid docs/blog"
    }
   ],
   "auth": [
    {
     "text": "Liquid API key (prefixed liquid_, created in console.liquid.ai under API Keys) as Authorization: Bearer $LIQUID_API_KEY.",
     "verification": "Verified 2026-10-16 on docs.liquid.ai/lfm/models/d1"
    }
   ],
   "fit": [
    "A free text-only tier (d1:free) is useful for prototyping. Liquid’s sources now agree on $0.02 per 1M input."
   ],
   "sources": [
    {
     "title": "Liquid docs: d1",
     "url": "https://docs.liquid.ai/lfm/models/d1"
    },
    {
     "title": "Liquid docs: Decision Models",
     "url": "https://docs.liquid.ai/lfm/models/decision-models"
    },
    {
     "title": "Liquid blog: Introducing d1",
     "url": "https://www.liquid.ai/blog/d1-decision-model"
    },
    {
     "title": "Liquid pricing",
     "url": "https://www.liquid.ai/pricing"
    },
    {
     "title": "Vercel AI Gateway: d1",
     "url": "https://vercel.com/ai-gateway/models/d1"
    }
   ]
  },
  {
   "slug": "strands-decider-2b",
   "name": "Strands Decider 2B (open source)",
   "vendor": "Strands Agents (AWS)",
   "category": "Open-weight decision model",
   "filter": "decision",
   "verified": "2026-10-15",
   "summary": "Small open-source decision model (Qwen3.5-2B-Base + LoRA adapter + option-scoring head) for local routing, tool selection, guardrails, and evals. Apache-2.0, weights on Hugging Face.",
   "url": "https://agentapilist.com/apis/strands-decider-2b/",
   "cost_calculator": null,
   "priced": false,
   "not_comparable": "Open weights; you pay your own compute, which depends on your hardware. No per-request price exists.",
   "pricing": [
    {
     "text": "Free open weights (Apache-2.0); you pay your own compute.",
     "verification": "Verified 2026-10-15 on Hugging Face model card"
    }
   ],
   "limits": [
    {
     "text": "Depends on your hardware; no hosted SLA.",
     "verification": "Verified 2026-10-15 against official page"
    }
   ],
   "status": [
    {
     "text": "First released 2026-10-09; model card live on Hugging Face. Self-reported JevBench public: 180/231, Brier 0.280, ECE 0.072.",
     "verification": "Verified 2026-10-15 (R&D re-check, no change) on Hugging Face model card"
    },
    {
     "text": "Built-in HTTP server binds to 127.0.0.1 with no authentication; meant for local experiments only.",
     "verification": "Verified 2026-10-15 on Hugging Face model card"
    }
   ],
   "auth": [],
   "fit": [
    "Best option here for pinning, on-prem, and reproducibility: you own the exact weights.",
    "Small model: validate accuracy on harder tasks before relying on it."
   ],
   "sources": [
    {
     "title": "Hugging Face model card",
     "url": "https://huggingface.co/StrandsAgents/strands-decider-2B-qwen3.5-v1-2610"
    },
    {
     "title": "Strands blog announcement",
     "url": "https://strandsagents.com/blog/introducing-strands-decider/"
    }
   ]
  },
  {
   "slug": "aws-bedrock-guardrails",
   "name": "AWS Bedrock Guardrails",
   "vendor": "AWS",
   "category": "Policy guardrail",
   "filter": "guardrail",
   "verified": "2026-10-16",
   "summary": "Configurable guardrails: content filters, denied topics, PII handling, contextual grounding, and Automated Reasoning checks; returns block/allow plus assessments.",
   "url": "https://agentapilist.com/apis/aws-bedrock-guardrails/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/aws-bedrock-guardrails/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "Content filters (text) $0.15 per 1,000 text units; denied topics $0.15/1k; sensitive-info filters $0.10/1k; regex and word filters free; contextual grounding $0.10/1k; Automated Reasoning checks $0.17/1k per policy; image content filter $0.00075/image. A text unit is up to 1,000 characters.",
     "verification": "Verified 2026-10-15 on AWS Bedrock pricing page"
    },
    {
     "text": "InvokeGuardrailChecks (checks-only API): content filters $0.07/1k text units, prompt attack $0.08/1k, sensitive info $0.10/1k.",
     "verification": "Verified 2026-10-15 on AWS Bedrock pricing page"
    }
   ],
   "limits": [
    {
     "text": "Default quotas (us-east-1, adjustable): 100 ApplyGuardrail requests per second; content filters 200 text units per second; content policy maximum input 1,000 text units per request.",
     "verification": "Verified 2026-10-16 on https://docs.aws.amazon.com/general/latest/gr/bedrock.html"
    }
   ],
   "status": [
    {
     "text": "Amazon Bedrock SLA: AWS commits to 99.9% monthly uptime per region, with service credits below that. The SLA page covers Amazon Bedrock as a whole and does not name Guardrails separately.",
     "verification": "Verified 2026-10-16 on aws.amazon.com/bedrock/sla"
    }
   ],
   "auth": [
    {
     "text": "AWS Signature Version 4 with IAM credentials (usually via the AWS SDK). Our unauthenticated test POST to bedrock-runtime ApplyGuardrail returned HTTP 403 “Authorization header is missing”.",
     "verification": "Verified 2026-10-16 on https://docs.aws.amazon.com/bedrock/latest/APIReference/CommonParameters.html and by test request to bedrock-runtime.us-east-1.amazonaws.com"
    }
   ],
   "fit": [
    "Only vendor feature in this tracker aimed at formal-logic verification (Automated Reasoning checks).",
    "Same pattern applies: deterministic rules own hard constraints; the model is advisory; fail closed on errors."
   ],
   "sources": [
    {
     "title": "AWS Bedrock pricing",
     "url": "https://aws.amazon.com/bedrock/pricing/"
    },
    {
     "title": "Bedrock Guardrails user guide",
     "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html"
    },
    {
     "title": "Amazon Bedrock SLA",
     "url": "https://aws.amazon.com/bedrock/sla/"
    }
   ]
  },
  {
   "slug": "azure-ai-content-safety",
   "name": "Content Safety in Foundry Control Plane (formerly Azure AI Content Safety)",
   "vendor": "Microsoft",
   "category": "Harm classifier",
   "filter": "classifier",
   "verified": "2026-10-16",
   "summary": "Microsoft harm classifiers with severity levels, Prompt Shields, and groundedness detection, now marketed as Content Safety in Foundry Control Plane.",
   "url": "https://agentapilist.com/apis/azure-ai-content-safety/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/azure-ai-content-safety/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "Free (F0) tier: 5,000 text records/month and 5,000 images/month; usage stops at the limit. A text record is up to 1,000 characters.",
     "verification": "Verified 2026-10-15 on Azure pricing page"
    },
    {
     "text": "Standard (S0) tier: $0.375 per 1,000 text records and $0.75 per 1,000 images (East US retail prices). A text record is up to 1,000 characters.",
     "verification": "Verified 2026-10-16 on https://prices.azure.com/api/retail/prices (Content Safety meters) and https://azure.microsoft.com/en-us/pricing/details/content-safety/"
    }
   ],
   "limits": [
    {
     "text": "Query rate limits: F0 (free) tier 5 requests/second for text and image moderation and Prompt Shields; S0 tier 1,000 requests per 10 seconds. Faster rates on request.",
     "verification": "Verified 2026-10-16 on learn.microsoft.com/azure/ai-services/content-safety/overview"
    }
   ],
   "status": [
    {
     "text": "Listed on the Azure pricing page under the new name.",
     "verification": "Verified 2026-10-15 on Azure pricing page"
    }
   ],
   "auth": [],
   "fit": [
    "Good as a content gate; not a general decision model.",
    "Same pattern applies: deterministic rules own hard constraints; the model is advisory; fail closed on errors."
   ],
   "sources": [
    {
     "title": "Azure pricing: Content Safety",
     "url": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/content-safety/"
    },
    {
     "title": "Microsoft Learn: Content Safety",
     "url": "https://learn.microsoft.com/azure/ai-services/content-safety/"
    },
    {
     "title": "Azure AI Content Safety overview (query rates)",
     "url": "https://learn.microsoft.com/en-us/azure/ai-services/content-safety/overview"
    }
   ]
  },
  {
   "slug": "google-vertex-structured-output",
   "name": "Gemini structured output (Vertex AI / Agent Platform)",
   "vendor": "Google Cloud",
   "category": "Structured output",
   "filter": "structured",
   "verified": "2026-10-16",
   "summary": "Constrain Gemini output to a JSON schema or a single enum value (response_mime_type text/x.enum) with response_schema. Google Cloud now documents this under the Gemini Enterprise Agent Platform name. No probabilities are documented on the structured-output page.",
   "url": "https://agentapilist.com/apis/google-vertex-structured-output/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/google-vertex-structured-output/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "Per-token Gemini pricing; for example Gemini 3.8 Flash is $0.75 / $3.75 per 1M input/output tokens (introductory, through 2026-12-31), then $1.50 / $7.50 from 2027-01-01. The response schema counts toward the input token limit.",
     "verification": "Verified 2026-10-16 on https://cloud.google.com/vertex-ai/generative-ai/pricing and the structured output page"
    }
   ],
   "limits": [
    {
     "text": "Response schema size counts toward the input token limit; schema must be defined via the API (no console support).",
     "verification": "Verified 2026-10-16 on Google Cloud structured output page (last updated 2026-10-15)"
    }
   ],
   "status": [
    {
     "text": "Supported models listed include Gemini 4 Argon, Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash, 3.5 Flash and Flash-Lite, and Gemini 2.5 Pro/Flash.",
     "verification": "Verified 2026-10-16 on Google Cloud structured output page"
    },
    {
     "text": "Correction: an earlier version of this page said you could read logprobs alongside the enum. The structured output page does not mention logprobs, so treat probabilities as unverified for this path.",
     "verification": "Correction logged 2026-10-16"
    }
   ],
   "auth": [
    {
     "text": "Google Cloud credentials: Vertex docs offer an API key or Application Default Credentials (gcloud auth).",
     "verification": "Verified 2026-10-16 on cloud.google.com Vertex AI docs (Get an API key / Configure application default credentials)"
    }
   ],
   "fit": [
    "text/x.enum approximates a choice API but returns one label with no per-option probability on the page we checked; calibration is yours to build.",
    "Same pattern applies: deterministic rules own hard constraints; the model is advisory; fail closed on errors."
   ],
   "sources": [
    {
     "title": "Google Cloud: Structured output",
     "url": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/control-generated-output"
    }
   ]
  },
  {
   "slug": "anthropic-claude-structured-output",
   "name": "Anthropic Claude (tool use / structured outputs)",
   "vendor": "Anthropic",
   "category": "Structured output",
   "filter": "structured",
   "verified": "2026-10-16",
   "summary": "JSON outputs (output_config.format) and strict tool use constrain Claude to a JSON schema via constrained decoding. Returns labels, not probabilities. Available via Anthropic, Bedrock, Google Cloud, and Microsoft Foundry.",
   "url": "https://agentapilist.com/apis/anthropic-claude-structured-output/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/anthropic-claude-structured-output/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "Per 1M input/output tokens: Haiku 5.5 $0.10/$0.50 (prompts up to 100k); Haiku 4.5 $1/$5; Sonnet 5.5 $2/$10; Opus 5.5 $4/$20. US-only inference_geo costs 1.1x; Batch is 50% off.",
     "verification": "Verified 2026-10-16 on Anthropic pricing docs"
    },
    {
     "text": "Structured outputs add an injected system prompt, so input tokens are slightly higher; changing the format invalidates the prompt cache.",
     "verification": "Verified 2026-10-16 on Anthropic structured outputs docs"
    }
   ],
   "limits": [
    {
     "text": "Per request: up to 20 strict tools, 24 optional parameters, and 16 union-type parameters across all strict schemas; grammar compilation timeout 180 s. Unsupported: recursive schemas, numeric/string length constraints, minItems other than 0 or 1.",
     "verification": "Verified 2026-10-16 on Anthropic structured outputs docs"
    },
    {
     "text": "Compiled grammars are cached for 24 hours from last use; the first request with a new schema has extra latency.",
     "verification": "Verified 2026-10-16 on Anthropic structured outputs docs"
    }
   ],
   "status": [
    {
     "text": "Generally available. Refusals (stop_reason \"refusal\") and max_tokens cut-offs can return output that does not match the schema; enum capitalization is not guaranteed.",
     "verification": "Verified 2026-10-16 on Anthropic structured outputs docs"
    },
    {
     "text": "No token probabilities are documented on the structured outputs page.",
     "verification": "Not published by vendor (checked 2026-10-16 on https://platform.claude.com/docs/en/build-with-claude/structured-outputs)"
    }
   ],
   "auth": [
    {
     "text": "Anthropic API key in the x-api-key header ($ANTHROPIC_API_KEY).",
     "verification": "Verified 2026-10-16 on docs.anthropic.com structured outputs page"
    }
   ],
   "fit": [
    "No probabilities means thresholds need another signal; a self-reported \"confidence\" field in your schema is generated text, not a calibrated probability.",
    "Compare enum values case-insensitively (Anthropic’s own caveat).",
    "Same pattern applies: deterministic rules own hard constraints; the model is advisory; fail closed on errors."
   ],
   "sources": [
    {
     "title": "Anthropic pricing",
     "url": "https://docs.anthropic.com/en/docs/about-claude/pricing"
    },
    {
     "title": "Anthropic structured outputs",
     "url": "https://docs.anthropic.com/en/docs/build-with-claude/structured-outputs"
    }
   ]
  },
  {
   "slug": "openai-moderation",
   "name": "OpenAI Moderation (omni-moderation)",
   "vendor": "OpenAI",
   "category": "Harm classifier",
   "filter": "classifier",
   "verified": "2026-10-16",
   "summary": "Free harm classifier returning per-category scores for text and images. Only the continuously upgraded omni-moderation-latest alias is documented.",
   "url": "https://agentapilist.com/apis/openai-moderation/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/?apis=openai-moderation",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "Free.",
     "verification": "Verified 2026-10-15 on OpenAI moderation guide"
    }
   ],
   "limits": [
    {
     "text": "Image files up to 20 MB; no audio.",
     "verification": "Verified 2026-10-15 on OpenAI moderation guide"
    }
   ],
   "status": [
    {
     "text": "Correction: only omni-moderation-latest is documented, and OpenAI says it plans to continuously upgrade the underlying model, so treat it as an unpinned alias. An earlier version of this page said \"dated snapshot\".",
     "verification": "Verified 2026-10-15 on OpenAI moderation guide"
    }
   ],
   "auth": [
    {
     "text": "OpenAI API key as Authorization: Bearer $OPENAI_API_KEY on POST /v1/moderations.",
     "verification": "Verified 2026-10-16 on platform.openai.com/docs/guides/moderation"
    }
   ],
   "fit": [
    "Content-safety gate only; not a general decision model.",
    "OpenAI warns category_scores thresholds may need recalibration as the model is upgraded; re-check thresholds periodically."
   ],
   "sources": [
    {
     "title": "OpenAI moderation guide",
     "url": "https://platform.openai.com/docs/guides/moderation"
    }
   ]
  },
  {
   "slug": "cohere-classify",
   "name": "Cohere Classify",
   "vendor": "Cohere",
   "category": "Classifier",
   "filter": "classifier",
   "verified": "2026-10-16",
   "summary": "Few-shot / fine-tuned classification at POST /v1/classify returning per-label confidences. v1 path only; there is no v2 classify.",
   "url": "https://agentapilist.com/apis/cohere-classify/",
   "cost_calculator": null,
   "priced": false,
   "not_comparable": "No Classify price is published on cohere.com/pricing.",
   "pricing": [
    {
     "text": "No Classify price found. cohere.com/pricing does not list Classify (it lists Model Vault instances and legacy Command/Aya token prices). Trial keys are free but rate limited and not for production.",
     "verification": "Not published by vendor (checked 2026-10-16 on https://cohere.com/pricing)"
    }
   ],
   "limits": [
    {
     "text": "Up to 96 inputs per call; 2–2,500 inline examples.",
     "verification": "Verified 2026-10-15 on Cohere API reference"
    }
   ],
   "status": [
    {
     "text": "Deprecated: Cohere lists /v1/classify among endpoints deprecated effective 2025-09-15. It is still documented on the v1 path only (no v2 classify) and still answers requests (our unauthenticated test got HTTP 401).",
     "verification": "Verified 2026-10-16 on https://docs.cohere.com/docs/deprecations and https://docs.cohere.com/reference/classify"
    }
   ],
   "auth": [
    {
     "text": "Bearer token: Authorization: Bearer <token>, where the token is your Cohere API key.",
     "verification": "Verified 2026-10-16 on docs.cohere.com/reference/classify"
    }
   ],
   "fit": [
    "Deprecated since 2025-09-15 with no listed price; do not build new work on it.",
    "Same pattern applies: deterministic rules own hard constraints; the model is advisory; fail closed on errors."
   ],
   "sources": [
    {
     "title": "Cohere API reference: Classify",
     "url": "https://docs.cohere.com/reference/classify"
    },
    {
     "title": "Cohere pricing",
     "url": "https://cohere.com/pricing"
    }
   ]
  },
  {
   "slug": "laya",
   "name": "Laya (open source)",
   "vendor": "Convai Innovations",
   "category": "Open-weight decision model",
   "filter": "decision",
   "verified": "2026-10-16",
   "summary": "Open-weight 421M encoder decision model (ModernBERT-large), Apache-2.0. laya-serve exposes a System One-style /v1/systemone endpoint; also runs in llama.cpp.",
   "url": "https://agentapilist.com/apis/laya/",
   "cost_calculator": null,
   "priced": false,
   "not_comparable": "Open weights; you pay your own compute, which depends on your hardware. No per-request price exists.",
   "pricing": [
    {
     "text": "Free open weights (Apache-2.0); you pay your own compute.",
     "verification": "Verified 2026-10-15 on Hugging Face model card"
    }
   ],
   "limits": [
    {
     "text": "512-token context.",
     "verification": "Verified 2026-10-15 on Hugging Face model card"
    }
   ],
   "status": [
    {
     "text": "Hugging Face model and GitHub repo were both created on 2026-09-18 (UTC), which matches the third-party release date. Latest GitHub release: v0.4.8, published 2026-10-15.",
     "verification": "Verified 2026-10-16 on https://huggingface.co/api/models/convaiinnovations/laya and https://api.github.com/repos/NandhaKishorM/laya/releases"
    },
    {
     "text": "Our smoke test (script: /workspace/research/bench/laya_bench.py): local CPU on an 8-vCPU box, median 181 ms and p95 245 ms per call, 10/10 correct on easy synthetic items, stable across 5 repeats and reversed option order. Small smoke test only, not a quality benchmark. Confidences are uncalibrated (the loader warns the checkpoint ships invalid temperatures).",
     "verification": "Verified 2026-10-15 — AgentAPIList local run"
    }
   ],
   "auth": [
    {
     "text": "Self-hosted. laya-serve has no authentication unless LAYA_API_KEY is set; then it requires Authorization: Bearer <key>.",
     "verification": "Verified 2026-10-16 on huggingface.co/convaiinnovations/laya"
    }
   ],
   "fit": [
    "Tiny and fast enough for CPU routing and gating; recalibrate before using confidence thresholds.",
    "Short 512-token context; truncate or summarize long state."
   ],
   "sources": [
    {
     "title": "Hugging Face model card",
     "url": "https://huggingface.co/convaiinnovations/laya"
    },
    {
     "title": "GitHub: laya",
     "url": "https://github.com/NandhaKishorM/laya"
    }
   ]
  },
  {
   "slug": "llama-cpp-decision-endpoint",
   "name": "llama.cpp decision endpoint (/v1/systemone)",
   "vendor": "ggml-org (llama.cpp)",
   "category": "Open-weight decision model",
   "filter": "decision",
   "verified": "2026-10-16",
   "summary": "llama-server in llama.cpp now serves a local System One-style /v1/systemone endpoint for open decision models.",
   "url": "https://agentapilist.com/apis/llama-cpp-decision-endpoint/",
   "cost_calculator": null,
   "priced": false,
   "not_comparable": "Free open-source server; compute cost depends on your hardware.",
   "pricing": [
    {
     "text": "Free and open source; you pay your own compute. Check each model’s license (OpenJev is CC BY-NC 4.0, non-commercial).",
     "verification": "Verified 2026-10-15 on Hugging Face blog"
    }
   ],
   "limits": [
    {
     "text": "Depends on your hardware and the model; no hosted SLA.",
     "verification": "Verified 2026-10-15 on Hugging Face blog"
    }
   ],
   "status": [
    {
     "text": "Published Oct 2, 2026 on the Hugging Face blog by ggml-org.",
     "verification": "Verified 2026-10-15 on Hugging Face blog"
    }
   ],
   "auth": [
    {
     "text": "llama-server supports optional API key auth with --api-key (default: none). The announcement post itself does not describe authentication.",
     "verification": "Verified 2026-10-16 on https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md"
    }
   ],
   "fit": [
    "Run decision models on-prem behind the same request shape as hosted APIs.",
    "Confirm licenses before commercial use."
   ],
   "sources": [
    {
     "title": "Hugging Face blog: Decision models in llama.cpp",
     "url": "https://huggingface.co/blog/ggml-org/decision-models-in-llamacpp"
    }
   ]
  },
  {
   "slug": "deepseek-search-fetch",
   "name": "DeepSeek Search & Fetch API",
   "vendor": "DeepSeek",
   "category": "Web search & fetch (agent tool)",
   "filter": "search",
   "verified": "2026-10-17",
   "summary": "Two paid web-access endpoints for agents and RAG: Search returns ranked results from DeepSeek’s own search engine, and Fetch turns one URL into Markdown. $0.75 per 1,000 successful calls on each, billed per call rather than per token. Opened to all API users on Oct 17, 2026.",
   "url": "https://agentapilist.com/apis/deepseek-search-fetch/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/?apis=deepseek-search-fetch",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "Search $0.75 and Fetch $0.75 per 1,000 successful requests. Only HTTP 200 responses with a result are billed: empty searches, failed fetches and timeouts are free, and count does not change the price. Charges come out of your topped-up or granted balance.",
     "verification": "Verified 2026-10-17 on https://api-docs.deepseek.com/quick_start/pricing and https://api-docs.deepseek.com/guides/search"
    },
    {
     "text": "Search and Fetch calls made as server-side tools through DeepSeek Harness, Claude Code or Codex are free until per-request billing starts at 16:00 UTC on Oct 31, 2026 (00:00 Beijing time, Nov 1). There is no free tier for direct calls.",
     "verification": "Verified 2026-10-17 on https://api-docs.deepseek.com/guides/search and https://api-docs.deepseek.com/news/news261017"
    }
   ],
   "limits": [
    {
     "text": "5 requests per second for Search and 5 for Fetch, per account across all keys, counted separately and apart from model concurrency limits. Fetch also has a per-domain limit (number not published). Over-limit requests get HTTP 429.",
     "verification": "Verified 2026-10-17 on https://api-docs.deepseek.com/guides/search"
    },
    {
     "text": "Search query 1–128 characters and 1–50 results per call; Fetch URL up to 8,192 characters. Oversized bodies return 413.",
     "verification": "Verified 2026-10-17 on https://api-docs.deepseek.com/api/search and https://api-docs.deepseek.com/api/fetch"
    }
   ],
   "status": [
    {
     "text": "Opened to everyone on Oct 17, 2026 (DeepSeek changelog and news post). The docs and news post attach no early-access or test-release label, and publish no SLA; a status page is at status.deepseek.com.",
     "verification": "Verified 2026-10-17 on https://api-docs.deepseek.com/updates and https://api-docs.deepseek.com/news/news261017"
    },
    {
     "text": "Data terms: the DeepSeek Open Platform Terms are governed by mainland PRC law, and DeepSeek’s privacy policy says it collects, processes and stores personal data in the People’s Republic of China. The Search & Fetch docs publish no data-retention, region or residency options.",
     "verification": "Verified 2026-10-17 on https://cdn.deepseek.com/policies/en-US/deepseek-open-platform-terms-of-service.html and https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html"
    }
   ],
   "auth": [
    {
     "text": "DeepSeek API key as Authorization: Bearer <key> with Content-Type: application/json, the same key used for DeepSeek’s model API.",
     "verification": "Verified 2026-10-17 on https://api-docs.deepseek.com/guides/search"
    }
   ],
   "fit": [
    "Cheap flat per-call pricing for agent web access, and Search can return page content inline (need_content), which can save a separate Fetch call.",
    "5 requests per second per endpoint is low for crawling or high-concurrency agents.",
    "Data is processed under PRC terms with no residency option, which may rule it out for regulated workloads."
   ],
   "sources": [
    {
     "title": "DeepSeek docs: Search & Fetch guide",
     "url": "https://api-docs.deepseek.com/guides/search"
    },
    {
     "title": "DeepSeek API reference: Search",
     "url": "https://api-docs.deepseek.com/api/search"
    },
    {
     "title": "DeepSeek API reference: Fetch",
     "url": "https://api-docs.deepseek.com/api/fetch"
    },
    {
     "title": "DeepSeek Models & Pricing",
     "url": "https://api-docs.deepseek.com/quick_start/pricing"
    },
    {
     "title": "DeepSeek change log (2026-10-17)",
     "url": "https://api-docs.deepseek.com/updates"
    },
    {
     "title": "DeepSeek Claude Code integration (web search)",
     "url": "https://api-docs.deepseek.com/quick_start/agent_integrations/claude_code"
    },
    {
     "title": "DeepSeek Codex integration",
     "url": "https://api-docs.deepseek.com/quick_start/agent_integrations/codex"
    },
    {
     "title": "DeepSeek Open Platform Terms",
     "url": "https://cdn.deepseek.com/policies/en-US/deepseek-open-platform-terms-of-service.html"
    }
   ]
  },
  {
   "slug": "liquid-ai-d1-ultra",
   "name": "Liquid AI d1-ultra",
   "vendor": "Liquid AI",
   "category": "Decision model",
   "filter": "decision",
   "verified": "2026-10-17",
   "summary": "Liquid AI’s larger multimodal decision model, released Oct 15, 2026. It answers yes/no (noul), choice and score questions about text and images with calibrated probabilities and generates no tokens. $0.042 per 1M input tokens, output free. Top decision model on the Artificial Analysis Decision Index (65.2, checked 2026-10-17).",
   "url": "https://agentapilist.com/apis/liquid-ai-d1-ultra/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/liquid-ai-d1-ultra/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "$0.042 per 1M input tokens, cached input $0.042, output $0. Liquid says the price is the same on Liquid Console, OpenRouter and Vercel AI Gateway; new Liquid Console organizations get $2 free credit.",
     "verification": "Verified 2026-10-17 on https://www.liquid.ai/pricing"
    },
    {
     "text": "Images are input tokens at 1.5 tokens per 32×32-pixel patch (a 1024×1024 image is 1,536 tokens). Each question is billed separately for its text and all images, so several questions about one image multiply the image cost.",
     "verification": "Verified 2026-10-17 on https://www.liquid.ai/blog/d1-ultra and https://docs.liquid.ai/lfm/models/d1"
    }
   ],
   "limits": [
    {
     "text": "Context window 32,768 tokens, as listed by Vercel AI Gateway and OpenRouter. Liquid’s own docs do not state a context window.",
     "verification": "Verified 2026-10-17 on https://vercel.com/changelog/liquid-ais-d1-ultra-now-available-on-ai-gateway and https://openrouter.ai/liquid/d1-ultra"
    },
    {
     "text": "Images: up to 8 per request, JSON body under 4.5 MB, up to 10,000 32×32 patches across all images, JPEG/PNG/WebP/GIF as base64 (remote image URLs are not accepted). Rate limits are not published.",
     "verification": "Verified 2026-10-17 on https://docs.liquid.ai/lfm/models/d1"
    }
   ],
   "status": [
    {
     "text": "Released Oct 15, 2026 on the Liquid API. Vercel AI Gateway announced availability the same day and OpenRouter lists it (release date Oct 15, 2026), although Liquid’s launch post said those two were coming “soon”. No release-stage label or SLA is stated.",
     "verification": "Verified 2026-10-17 on https://www.liquid.ai/blog/d1-ultra, https://vercel.com/changelog/liquid-ais-d1-ultra-now-available-on-ai-gateway and https://openrouter.ai/liquid/d1-ultra"
    }
   ],
   "auth": [
    {
     "text": "Liquid API key (prefixed liquid_, from console.liquid.ai) as Authorization: Bearer $LIQUID_API_KEY, the same key as d1. On OpenRouter and Vercel AI Gateway you use that platform’s own key with the model name liquid/d1-ultra.",
     "verification": "Verified 2026-10-17 on https://docs.liquid.ai/lfm/models/d1 and https://vercel.com/changelog/liquid-ais-d1-ultra-now-available-on-ai-gateway"
    }
   ],
   "fit": [
    "Strongest public benchmark result among hosted decision models right now; still validate on your own labeled data.",
    "Costs about twice d1 per input token, and each question is billed for every image again, so ask fewer questions per image where you can.",
    "Context is 32,768 tokens, far shorter than Perplexity’s 256K; long documents need chunking."
   ],
   "sources": [
    {
     "title": "Liquid blog: d1-ultra launch",
     "url": "https://www.liquid.ai/blog/d1-ultra"
    },
    {
     "title": "Liquid docs: d1 and d1-ultra",
     "url": "https://docs.liquid.ai/lfm/models/d1"
    },
    {
     "title": "Liquid pricing",
     "url": "https://www.liquid.ai/pricing"
    },
    {
     "title": "Vercel changelog: d1-ultra on AI Gateway",
     "url": "https://vercel.com/changelog/liquid-ais-d1-ultra-now-available-on-ai-gateway"
    },
    {
     "title": "OpenRouter: liquid/d1-ultra",
     "url": "https://openrouter.ai/liquid/d1-ultra"
    },
    {
     "title": "Artificial Analysis Decision Index",
     "url": "https://artificialanalysis.ai/evaluations/artificial-analysis-decision-index"
    }
   ]
  },
  {
   "slug": "h2o-lightning-4b",
   "name": "H2O-Lightning-4B (open source)",
   "vendor": "H2O.ai",
   "category": "Open-weight decision model",
   "filter": "decision",
   "verified": "2026-10-16",
   "summary": "Open-weight 4B decision model from H2O.ai (built on Qwen3.5-4B), Apache-2.0. Runs on stock vLLM 0.30.0 behind a small shim that serves /v1/systemone and an OpenAI-compatible /v1/decisions route. Text and up to 4 images per request.",
   "url": "https://agentapilist.com/apis/h2o-lightning-4b/",
   "cost_calculator": null,
   "priced": false,
   "not_comparable": "Open weights; the $0.021 per 1,000 decisions figure is H2O’s own self-hosting estimate, not a price.",
   "pricing": [
    {
     "text": "Free open weights (Apache-2.0); you pay your own GPU. H2O estimates $0.021 per 1,000 decisions self-hosted (its own estimate).",
     "verification": "Verified 2026-10-16 on Hugging Face model card (cost figure is a vendor estimate)"
    }
   ],
   "limits": [
    {
     "text": "Up to 4 images per request, 20 MB each, scaled to at most 1,048,576 pixels; 40,960-token max model length as shipped; up to 255 options (H2O says it decides best with about 16 or fewer). Over-limit inputs get HTTP 422. Weights are 9.3 GB.",
     "verification": "Verified 2026-10-16 on Hugging Face model card"
    }
   ],
   "status": [
    {
     "text": "H2O reports v1.1 was #1 on the JevBench open-weight composite from Oct 7 to Oct 16, 2026 (72.5 vs 71.5 for hosted Jev 1.13.0).",
     "verification": "Unverified vendor self-report: the current third-party leaderboard does not show it (see below; checked 2026-10-16)"
    },
    {
     "text": "H2O’s own runs: 205/231 on JevBench’s public items (v1.2.3 weights); 202/231 for v1.3 two-way. Serial client latency p50 about 29–30 ms on one H100 or RTX PRO 4500.",
     "verification": "Verified 2026-10-16 as published on the model card (vendor-measured, not reproduced by us)"
    }
   ],
   "auth": [
    {
     "text": "Open weights you run yourself. The model card’s local server example ignores the API key (api_key=\"unused\").",
     "verification": "Verified 2026-10-16 on huggingface.co/h2oai/h2o-lightning-4b"
    }
   ],
   "fit": [
    "Strong option for self-hosted gating and routing with images, if you have a GPU; mostly evaluated in English.",
    "Yes/no answers are floored at 0.801 by default; turn the floor off (SHIM_NOUL_FLOOR=0) if you escalate uncertain cases.",
    "Not a chat model: no explanations."
   ],
   "sources": [
    {
     "title": "Hugging Face model card",
     "url": "https://huggingface.co/h2oai/h2o-lightning-4b"
    },
    {
     "title": "H2O press release",
     "url": "https://www.h2o.ai/company/press-media/2026/h2o-lightning-4b-ranks-first-on-jevbench"
    }
   ]
  },
  {
   "slug": "llm-gateway-decisions",
   "name": "LLM Gateway — Decisions",
   "vendor": "LLM Gateway",
   "category": "Gateway / proxy",
   "filter": "gateway",
   "verified": "2026-10-16",
   "summary": "Gateway with a System One endpoint (/v1/systemone, launched Sep 20, 2026) and an OpenAI-compatible /v1/decisions endpoint (added Oct 12, 2026) serving Jev, GPT-6 Luna Decisions, and Microsoft-Decision-1.",
   "url": "https://agentapilist.com/apis/llm-gateway-decisions/",
   "cost_calculator": "https://agentapilist.com/cost-calculator/llm-gateway-decisions/",
   "priced": true,
   "not_comparable": null,
   "pricing": [
    {
     "text": "Jev 1.13: $0.042 per 1M input tokens, output free. Prices for the other decision models on the gateway were not checked.",
     "verification": "Verified 2026-10-16 on LLM Gateway changelog"
    }
   ],
   "limits": [
    {
     "text": "Per-organization rate limit of 600 requests per minute on /v1/systemone; choice up to 255 options; score 2–10 levels.",
     "verification": "Verified 2026-10-16 on LLM Gateway changelog"
    }
   ],
   "status": [
    {
     "text": "/v1/decisions added Oct 12, 2026; /v1/systemone keeps working for TypeSafe models but “will be deprecated soon”.",
     "verification": "Verified 2026-10-16 on LLM Gateway changelog"
    }
   ],
   "auth": [
    {
     "text": "LLM Gateway API key as Authorization: Bearer $LLM_GATEWAY_API_KEY on POST https://api.llmgateway.io/v1/systemone.",
     "verification": "Verified 2026-10-16 on llmgateway.io/changelog/system-one-typed-decisions"
    }
   ],
   "fit": [
    "Use /v1/decisions for new code, since /v1/systemone is slated for deprecation.",
    "Adds a third party in the data path; check its compliance policies against your requirements."
   ],
   "sources": [
    {
     "title": "LLM Gateway changelog: System One typed decisions",
     "url": "https://llmgateway.io/changelog/system-one-typed-decisions"
    }
   ]
  }
 ]
}