Nace AI launches Drex Max, a screenshot-in decision model for agents at 174 ms and $0.06/M tokens
testingcatalog · x · 2026-10-19
Nace AI released Drex Max, a multimodal "decision model" for computer-use and web agents: it takes screenshots, images or video and returns calibrated probability scores over a fixed set of actions instead of free text. Claimed specs: decisions in 34-174 ms, 262K-token context, $0.06 per million input tokens — up to 167× cheaper per token than the Claude/GPT/Gemini models used for computer use. Vendor benchmarks show OSWorld-Verified 0.780, BFCL 98.2, median latency 183 ms and $0.06 per 1K decisions, vs 2-3.5 s and $3.33-$10.17 for frontier LLMs. The Drex family also includes open-weights Drex 2.0 for local serving; demos include real-browser flight booking, highway CCTV crash detection and per-tick decisions in StarCraft II and Doom.
More from coding & agent
- Blender renders cut from 4 hours to minutes with Opus 5.5 and cloud GPUs — ravisparikh · 2026-10-19
- LangChain Hosts Life Sciences Agent Engineering Workshop: Build, Trace, Evaluate Research Agents — LangChain · 2026-10-19
- Librarian Agents: new paper lets agents curate their own long-context workspaces — charles_irl · 2026-10-19
- Delta, JPMorgan and Schwab turn outside AI agents away — 'ungoverned AI' is what gets blocked — alex_verem · 2026-10-19
- The 80/20 of using Claude: six features that replace six time-wasting habits — alex_verem · 2026-10-19
- Open-source ai-memory gives Claude Code, Codex, Cursor and 20+ AI coding tools a shared memory — alex_verem · 2026-10-19