Microsoft ships Decision-1 at $0.042 per million input
Microsoft put a decision model of its own on OpenRouter overnight, claiming it is 4.5x faster than the runner-up. Every accuracy and latency figure so far is Microsoft's, and nobody outside has run it against Jev.
Built with it
Harrison Chase says Open SWE routing cut median cost per task 64%
Harrison Chase says Open SWE moved model choice into the harness, sending each task to the cheapest model that still does the job, tested against quality, and median cost per task dropped 64%. He was agreeing with @yuhasbeentaken, who says DeepSeek v4.1 Flash now handles orchestration, repetitive implementation and verification, with Opus or Sol kept for harder edge cases.
most orchestration steps don't need a frontier model
Someone measured it
Microsoft ships Decision-1, OpenRouter lists it at $0.042 per million input
Satya Nadella announced Microsoft-Decision-1 and OpenRouter listed it at $0.042 per million input tokens, output free, 32K context. Microsoft's own numbers, relayed by OpenRouter, claim highest accuracy across 36 blind benchmarks of about 150K questions, 4.5x faster than the runner-up, and decisions flipping on 1.3% of perturbed inputs. It is post-trained from Qwen3.5-9B.
It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality.
The case against
Mike Taylor asks if Microsoft-Decision-1 is just constrained output tokens
Mike Taylor questioned whether Microsoft-Decision-1 is a decision model at all, replying to Lisan al Gaib's note on how fast everyone deployed decision models. Nadella's announcement says only that it delivers top performance on structured decision tasks, with no architecture detail; OpenRouter's listing says it was post-trained from Qwen3.5-9B.
Is it actually a decision model though? Or just constrained output tokens?