Chrome's new DecisionModel API reverse-engineered: prompts, limits and engine tests
dejanseo · x · 2026-10-08
Chrome quietly landed the DecisionModel API in Chromium (Oct 6-8, behind a desktop-only flag): it lets sites ask closed-ended questions about visitor text on-device and get per-answer probabilities. RESONEO found it first and tested three engines across 414 queries; Dan Petrovic digs deeper:
- Full prompts: extracts the complete Gemma 4 engine system prompt from decisionmodelpromptbuilder.cc — input treated only as data to classify, instructions inside ignored, reply is a single option letter; questions sent one at a time as A/B choices.
- Two engines in review: the article parses 6 open code reviews and 143 reviewer comments for limits and the math behind scoring.
- Local runs: benchmarks against Google's public MediaPipe DecisionMaker plus hands-on tests of EmbeddingGemma 1/2 and Laya.
More from Infra
- llama.cpp launch of Qwen3.8 MTP files on 2x DGX Spark cluster hits a dgx-stall — user seeks help — Impossible_Art9151 · 2026-10-08
- Samsung open-sources LittleBit, squeezing a 13B LLM under 1GB with 11.6x inference speedup — ChrSzegedy · 2026-10-08
- Six years on, A100 still leads Nvidia chip mentions in AI papers — ahead of H100+H200 — nathanbenaich · 2026-10-08
- Cloud backlog hits $1.69T as CoreWeave logs $2.58B quarterly revenue — nathanbenaich · 2026-10-08
- China's Power Glut Meets Data Centers; Immersion Cooling Traced to Bitcoin Miners — teortaxesTex · 2026-10-08
- omarchy-cluster runs the full 753B-param GLM-5.3 across four old Macs as one endpoint — natesiggard · 2026-10-08