Nace.AIdrex-max-v1.1 (2026-10-28)Verified 2026-10-30

Nace.AI Drex (Drex Max)

A decision API: send a state plus named choice, yes/no (noul) or score questions to POST https://console.nace.ai/v1/systemone and get a probability for every option back, with nothing generated. Drex Max (drex-max, released 2026-10-13) and Drex Max v1.1 (drex-max-v1.1, 2026-10-28) read text and images at $0.03 per 1M input tokens; output tokens are not billed. Drex Max v1 weights are also on Hugging Face under a modified Open RAIL-M license.1

CompareDocs

PricingUSD, as published

ItemValue
List price$0.03/1M in2
Free tierYes

Drex Max and Drex Max v1.1: $0.03 per 1M input tokens. Output tokens are not billed. No per-request or per-question fee. Drex 2.0 is $0.06 and Drex 1.5 is $0.05 per 1M input tokens.2 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/reference/pricing]

Only HTTP 200 responses are charged; 401, 402, 422, 429, 500 and 529 cost nothing.2 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/reference/pricing]

New accounts get a $5.00 signup balance (expires after 3 months) and a $5 monthly allowance that does not carry over, subject to eligibility rules (one per mailbox, proof of a person such as Google sign-in or a verified card). Top-ups $5 to $1,000 by card; pay-as-you-go billing every 14 days is optional.2 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/reference/pricing]

Estimate cost at 10k to 10M requests, or set your own usage.

LimitsPer request

LimitValue
Context window262,144 tokens
Rate limit120 / 600 req/min

Per request: up to 512 questions and 255 options per choice question (or levels per score question); serialized body up to 4 MiB (4,194,304 bytes).43 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/guides/drex-max and https://console.nace.ai/docs/drex/reference/limits]

Context: drex-max-v1.1 allows 262,144 tokens for the whole packed request (state, images and all questions); drex-max allows 262,144 tokens per question (state plus that question).13 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/reference/models and https://console.nace.ai/docs/drex/reference/limits]

Rate limits per account: free tier 120 requests/min and 8 in flight; paid tier (after any top-up, auto-pay or paid invoice) 600 requests/min and 16 in flight. Over the limit returns 429.3 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/reference/limits]

Timeouts: 55 s upstream evaluate timeout (then 529 overloaded with retry-after-ms 2000) inside a 60 s route limit.3 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/reference/limits]

The Drex API reference documents one synchronous JSON response per POST /v1/systemone. It documents no streaming, no batch or async endpoint, no tool or function calling and no custom JSON-schema output: the body takes state, questions, model and (Drex Max) images, and the answer is a probability per option (Nace: Drex “does not return a verdict”). Batching is per request: up to 512 questions on one state.56 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/api-reference/systemone and https://console.nace.ai/docs/drex/guides/reading-answers]

Statusdrex-max-v1.1 (2026-10-28)

  • drex-max released 2026-10-13 and drex-max-v1.1 released 2026-10-28, both listed as active models; drex-latest still points to drex-v2.0, so send a Drex Max id explicitly.1 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/reference/models]
  • Weights license: Nace AI Open RAIL-M License (Modified), version 1.0. Commercial use is not licensed (a separate written license is required) if you or your employer had over $1,000,000 in gross revenue in the prior 12 months or raised over $1,000,000 in equity or debt funding, except personal or research use, or if you offer any product competing with Nace’s, including decision models and document intelligence. It also bars fully automated decisions with legal or similarly significant effects on people without meaningful human review. It permits derivatives, including fine-tuned and quantized models, under the same terms.7 [Verified 2026-10-30 on https://huggingface.co/nace-ai/drex-max-v1/blob/main/LICENSE]
  • No SLA or uptime commitment is published in Nace’s Terms of Service or the Drex docs; when the model does not answer within 55 s the API returns 529 overloaded.83 [Verified 2026-10-30 on https://console.nace.ai/terms and https://console.nace.ai/docs/drex/reference/limits]
  • Data location: Nace says the services are operated from the United States and that it and its service providers process and store personal information in the US and other countries; no EU region is offered. Customer content is not used for training by default. The SDKs also document custom and on-prem deployments via NACE_BASE_URL.9810 [Verified 2026-10-30 on https://console.nace.ai/privacy, https://console.nace.ai/terms and https://console.nace.ai/docs/developer-tools/python]
  • Hosted fine-tuning: none documented in the Drex docs or SDKs; Nace’s terms say it does not train or fine-tune shared models on customer content unless you opt in.810 [Verified 2026-10-30 on https://console.nace.ai/terms and https://console.nace.ai/docs/developer-tools/python]

Auth

MethodAPI key (Bearer)
DetailsAPI key sent as Authorization: Bearer (the docs use $NACE_API_KEY). Up to 3 active keys per account.43 [Verified 2026-10-30 on https://console.nace.ai/docs/drex/guides/drex-max and https://console.nace.ai/docs/drex/reference/limits]

FeaturesFrom lib/apiFeatures.js

YesNoN/ANot published
CapabilityStateNote
Free tier$5 signup + $5/month balanceYes2$5 signup + $5/month balance
MCP supportHosted MCP (decide) + local nace-mcpYesHosted MCP (decide) + local nace-mcp
Fine-tuningOpen weights only; license allows fine-tuned derivativesYes7Open weights only; license allows fine-tuned derivatives
Official SDKsPython and TypeScript (nace-sdk), CLIYesPython and TypeScript (nace-sdk), CLI
Open source / self-hostDrex Max v1 weights, modified Open RAIL-MYes7Drex Max v1 weights, modified Open RAIL-M
Structured JSON outputTyped probabilities, no custom JSON schemaNo56Typed probabilities, no custom JSON schema
Tool / function callingDecision model: probabilities onlyNo56Decision model: probabilities only
EU / data residencyProcessed in the US; no EU regionNo9810Processed in the US; no EU region
BatchNo batch endpoint; up to 512 questions per requestNot published56No batch endpoint; up to 512 questions per request
StreamingOne synchronous JSON responseNot published56One synchronous JSON response
SLA publishedNo SLA in terms or docsNot published83No SLA in terms or docs

Fit and safety

Good for

  • Pin drex-max or drex-max-v1.1: omitting model, or sending drex-latest, gets drex-v2.0 at $0.06.
  • Read the license before self-hosting: an organization with over $1M in revenue or funding, or one offering a competing product, needs a separate commercial license from Nace for the open weights.

Watch out

  • Good for cheap, typed agent decisions (routing, yes/no gates, scoring) on long text or image state, at $0.03 per 1M input tokens with free output.
  • Benchmark scores on the model card (e.g. 64 on the Artificial Analysis Decision Index, 67.52 on Decision Index 0.4) are vendor-reported and not reproduced by us.

How it works

  1. Models (API id, released, price per 1M input tokens): drex-max-v1.1 (2026-10-28, $0.03), drex-max (2026-10-13, $0.03), drex-v2.0 (2026-10-06, $0.06, the default when model is omitted and the current drex-latest), drex-v1.5 (2026-09-28, $0.05).
  2. Question types: choice (up to 255 options), noul (yes/no) and score (up to 255 levels). Each answer carries probabilities over the options; the model card describes a 255-way readout head and one forward pass, with no generated text.
  3. Only drex-max and drex-max-v1.1 accept images: up to 8 per request, PNG, JPEG or WebP, 16 MiB each, 32 MiB in all, uploaded files or inline base64. Images are resized to at most 1,638,400 pixels and image tokens are billed as input tokens.
  4. drex-v1.0 was retired on 2026-10-21; requests that name it are answered by drex-max at drex-max's price.
  5. Open weights: nace-ai/drex-max-v1 on Hugging Face, which the card describes as a dense 27B-parameter multimodal model with a 262,144-token window (16,384 by default). It runs with Python on an NVIDIA GPU (about 49 GB in bf16, about 26 GB with the backbone in 8-bit); the text backbone also runs on Apple silicon or CPU via Nace forks of llama.cpp and Ollama, Kapitsa and MLX (Q8_0 GGUF and affine Q4 checkpoints).
  6. Official SDKs: nace-sdk for Python 3.10+ (pip install nace-sdk) and nace-sdk for TypeScript/Node 20+ (ESM and CommonJS), with system_one / systemOne and models.list; they keep the TypeSafe SDK’s names, so TypeSafe code ports by renaming imports. Nace also documents a nace CLI.
  7. MCP: a hosted MCP server at https://console.nace.ai/mcp (streamable HTTP) exposes decide and list_models, usable without an API key within free per-network limits and on your account with a key; a local MCP server (nace-mcp) exposes 53 tools across Drex, Perception and Workspaces.

Compare

Put Nace.AI Drex (Drex Max) side by side with other APIs on price, limits and features. Open the comparison table →

Deciding between a decision model and a generative LLM? See Decision model vs LLM: when to use which.

SourcesFootnotes

  1. 1Nace Drex docs: Modelsconsole.nace.ai/docs/drex/reference/modelsVerified 2026-10-30
  2. 2Nace Drex docs: Pricingconsole.nace.ai/docs/drex/reference/pricingVerified 2026-10-30
  3. 3Nace Drex docs: Limitsconsole.nace.ai/docs/drex/reference/limitsVerified 2026-10-30
  4. 4Nace Drex docs: Drex Max guideconsole.nace.ai/docs/drex/guides/drex-maxVerified 2026-10-30
  5. 5Nace docs: POST /v1/systemone referenceconsole.nace.ai/docs/drex/api-reference/systemoneVerified 2026-10-30
  6. 6Nace docs: Reading answersconsole.nace.ai/docs/drex/guides/reading-answersVerified 2026-10-30
  7. 7Hugging Face: nace-ai/drex-max-v1 LICENSEhuggingface.co/nace-ai/drex-max-v1/blob/main/LICENSEVerified 2026-10-30
  8. 8Nace Terms of Serviceconsole.nace.ai/termsVerified 2026-10-30
  9. 9Nace Privacy Noticeconsole.nace.ai/privacyVerified 2026-10-30
  10. 10Nace docs: Python SDKconsole.nace.ai/docs/developer-tools/pythonVerified 2026-10-30
  11. 11Hugging Face: nace-ai/drex-max-v1 model cardhuggingface.co/nace-ai/drex-max-v1Cited
  12. 12Nace docs: TypeScript SDKconsole.nace.ai/docs/developer-tools/typescriptCited
  13. 13Nace docs: Hosted MCP serverconsole.nace.ai/docs/developer-tools/hosted-mcpCited
  14. 14Nace docs: MCP serverconsole.nace.ai/docs/developer-tools/mcpCited