Microsoft Enters the Decision-Model Arena With Microsoft-Decision-1 Built on Qwen3.5-9B
Microsoft launched Microsoft-Decision-1 in Microsoft Foundry and on OpenRouter at Jev’s exact $0.042/M price—drawing headlines for basing its first release on Alibaba’s open Qwen3.5-9B.
Microsoft joined the post-Jev race on October 9–10 with the rollout of Microsoft-Decision-1 across Azure’s Microsoft Foundry and OpenRouter—matching TypeSafe AI’s $0.042 per million input token pricing with free output and offering Azure data-zone residency for enterprise buyers.
Pragmatic base-model choice draws industry attention
As Thomas Claburn reported in The Register on October 10, Microsoft’s rapid entry stood out for its underlying foundation: rather than waiting to train a proprietary model from scratch, Microsoft post-trained Alibaba’s open-weight Qwen3.5-9B for single-pass calibrated decision scoring over a 32K context window. Microsoft stated that future versions of its decision lineup will be rebased onto its own in-house MAI models as well as OpenAI models.
Same contract, enterprise Azure distribution
Like Cloudflare’s Clef, Microsoft-Decision-1 adopts the Jev-style structured choice, score, and yes/no probability contract, allowing teams to benchmark it side by side against Jev through OpenRouter or run multi-model editorial panels. For Azure-native enterprises with strict compliance boundaries, Microsoft’s move made governed System One inference immediately available without onboarding a new startup vendor.