Ordalin
Submit a tool

Reviewed tool profile

Drex Max

Nace.AI's 27B multimodal decision model that returns calibrated probabilities from screens and text.

Sources checked

Visit website ↗
Pricing
The hosted API bills input tokens only: Drex Max is $0.03 per million input tokens during launch, half its $0.06 list price, and output is not billed. New accounts get a $5 signup balance and a $5 monthly allowance; beyond that you top up by card or turn on pay as you go. Drex Max v1.1 is API-only (Nace Console and OpenRouter); the original Drex Max weights are downloadable from Hugging Face under Nace's modified Open RAIL-M licence. Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗
Available as
Hosted API on Nace Console (POST /v1/systemone), also listed by Nace on OpenRouter; the API is wire-compatible with Jev, so existing clients switch with three environment variables. · The original Drex Max weights run with Python on an NVIDIA GPU, with images; on Apple silicon and CPUs the text backbone runs through llama.cpp, Ollama, Kapitsa or MLX. · Nace also offers managed dedicated GPUs, deployment inside your own AWS or Azure account, or air-gapped in your datacenter. Official source 1 ↗ · Official source 2 ↗ Official source 1 ↗ Official source 1 ↗
Primary group
Automation & Agents

Also consider

See all 6 similar tools ↓

What it does

Drex Max is Nace.AI's multimodal decision model, first released on 13 October 2026 with a v1.1 update on 28 October. It does not write text: you send a case (text, JSON or up to 8 images such as screenshots, charts and forms) with typed questions, and one forward pass returns a calibrated probability for every option you supplied. Nace built it for AI agents that use computers and browsers, where each next click is a choice. It is a dense 27B model; Nace's technical report followed on 6 November 2026.

Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗ · Official source 4 ↗

Features

  • Probabilities, not prose. Each question is a choice between named options, a yes/no (noul) or a score on levels you define, and comes back as probabilities. It can only answer with options you gave it, so malformed JSON and invented options cannot occur. Official source 1 ↗ · Official source 2 ↗
  • Decides from screens and images. Drex Max reads screenshots, charts, forms, scanned documents, photos and video frames. Nace reports 98% accuracy on image-only questions from the raw image, against 63% from a text description, and it accepts screenshot sizes Jev 1.14 rejects. Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗
  • Built for computer-use agents. A computer-use agent makes dozens of screenshot decisions per task; Drex Max makes each in one pass. Nace's demos include buying a flight in a real browser, one choice per click. Official source 1 ↗
  • Calibrated confidence. Nace reports that when Drex Max is at least 93% sure it is right 97% of the time, and that reordering the options changes only 4.3% of answers. Official source 1 ↗
  • Many questions per request. Up to 512 questions and 8 images per request, with up to 255 options or levels per question; several questions about one case share a single pass. Official source 1 ↗ · Official source 2 ↗
  • Fast and cheap per decision. Nace quotes a 172 ms median per decision and $0.03 per thousand decisions at the launch price, against about 2 to 3.5 seconds for frontier chat models. Official source 1 ↗ · Official source 2 ↗
  • Leading decision benchmarks (vendor-reported). Nace reports 70.8 for v1.1 on the public Decision Index 0.4.0 against 60.7 for Jev 1.14, and its model card ranks the original Drex Max fourth of 18 on the Artificial Analysis Decision Index. Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗

Pricing & free limits

Free balance

$5Signup balance, then $5 monthly allowance

The signup balance expires after 3 months; the monthly allowance does not carry over. No card needed to try. Official source 1 ↗ · Official source 2 ↗

Drex Max API

$0.03 per 1M input tokensPay per use (launch price, 50% off $0.06 list)

Output tokens, per-request fees and failed requests are not billed. Top-ups $5 to $1,000 by card, or pay as you go billed every 14 days. Official source 1 ↗ · Official source 2 ↗

Open weights (Drex Max v1)

FreeDownload; Nace AI Open RAIL-M (modified)

Royalty-free for individuals and smaller organizations; companies over $1M revenue or funding, or competitors, need a commercial licence from Nace. Official source 1 ↗ · Official source 2 ↗

Billing details

Balances are spent first; without pay as you go, requests stop when the balance runs out. Top-ups are paid by card through Stripe in USD and expire 12 months after purchase. Official source 1 ↗

Only successful requests are billed; errors such as 422 or 529 cost nothing. Image tokens count as input tokens. Official source 1 ↗ · Official source 2 ↗

Browser runs, Nace's hosted browsing agent, are billed separately: $0.12 per million Drex Max input tokens plus planner tokens and $0.0045 per browser minute. Official source 1 ↗

Free access & limits

$5 signup balance (expires after 3 months) plus a $5 monthly allowance that does not carry over. Official source 1 ↗

Best for & limitations

Choosing an agent's next click or action from a screenshot during computer or browser use. Official source 1 ↗ · Official source 2 ↗

Routing support tickets and banking requests to the right team or intent with a probability you can threshold. Official source 1 ↗ · Official source 2 ↗

Reading financial news sentiment or contract clauses as typed decisions in finance and legal workflows. Official source 1 ↗ · Official source 2 ↗

Checking camera frames or images frame by frame, such as flagging crashes on highway CCTV. Official source 1 ↗

  • Restricted licence, not OSI open source. The Drex Max v1 weights need a separate commercial licence for companies with over $1M revenue or $1M funding (outside personal or research use) and for anyone offering a competing product; the restrictions carry over to derivatives. Official source 1 ↗
  • Latest version is API-only. Drex Max v1.1 has no open weights; it is offered only through Nace Console and OpenRouter. Official source 1 ↗
  • Use bans on automated decisions. The licence forbids fully automated decisions with legal or similar effects on people, such as in hiring, credit or insurance, without meaningful human review. Official source 1 ↗
  • Weaker on closed-book knowledge. Nace says it trails Jev 1.14 on MMLU-Pro, HLE and ARC, and names closed-book knowledge as its first remaining limitation. Official source 1 ↗ · Official source 2 ↗
  • Heavy to self-host. The bf16 weights take about 49 GB; on Apple silicon you need 24 GB or more for the Q4 build, and images only work on the Python GPU path. Official source 1 ↗
  • Image and request limits. Images are resized to about 1,280 × 1,280 pixels, at most 8 per request, and long image requests can be refused even when each question fits alone. Official source 1 ↗
  • Vendor-reported benchmarks. Speed, calibration and benchmark figures come from Nace's own pages and technical report. Official source 1 ↗ · Official source 2 ↗
Drex Max product screenshot
Website previewFixed desktop viewport · 1440 × 900

Similar tools

Published tools that share its categories.

FastinoOpen-weight decision models, led by GLiDE, that return probabilities over your own outcomes for agent routing, classification and scoring.
Automation & AgentsPaidChecked Nov 6
Visit
RerunNo-code AI coworkers that run your recurring business tasks on a schedule or trigger, on your own ChatGPT or Claude plan.
Automation & AgentsFree trialChecked Nov 6
Visit