Microsoft launches Decision-1, a probability-output decision model topping 36 blind benchmarks
usamawahabkhan · x · 2026-10-11
Microsoft-Decision-1 is live on OpenRouter: a small model post-trained from Qwen3.5-9B that outputs calibrated probabilities over fixed options instead of text, built for classification, routing, verification, workflow control, agent guardrails, and AI judging. Microsoft claims the highest accuracy across 36 blind benchmarks (150K questions), 4.5x faster than the runner-up, 35x faster than GPT-6 Sol, and only 1.3% decision flips on perturbed inputs. Priced at $0.042/M input tokens with free output and a 32K context; weights update continually while the API stays stable.
More from Models
- Local LLM field guide benchmarks 14 hardware configs for real-world token speed — NandoDF · 2026-10-11
- llama.cpp readies MiniCPM-V 4.7 support; hidden 35B-A3B model spotted before launch — jacek2023 · 2026-10-11
- AWS Bedrock sends EOL notices for Claude Sonnet 3.5, 3.7 and Haiku 3 — repligate · 2026-10-11
- Why buy $20k local machines for GLM 5.3 at 70 TPS? OpenRouter ran all night for $10 — TheZachMueller · 2026-10-11
- Dev who nearly burned a full week of usage finds Claude's limits more generous than expected — prasenx · 2026-10-11
- Musk says Grok autonomously bought LEGO from its website for him — elonmusk · 2026-10-11