Microsoft Launches Decision-1 at 4.2 Cents per Million Tokens, Claiming 35x GPT-6 Sol Speed
Updated
Updated · InfoWorld · Oct 13
Microsoft Launches Decision-1 at 4.2 Cents per Million Tokens, Claiming 35x GPT-6 Sol Speed
3 articles · Updated · InfoWorld · Oct 13
Summary
Decision-1 returns calibrated probabilities for predefined answers instead of free-form text, letting apps directly classify requests, prioritize incidents, verify AI outputs and route agent actions.
Microsoft says that design cuts cost and latency versus general-purpose LLMs: the model is priced at 4.2 cents per million input tokens, with $1 covering more than 275,000 sample incident-prioritization decisions.
On 36 benchmarks spanning nearly 150,000 blind questions, Microsoft said Decision-1 posted the highest accuracy, ran about 2.5 times faster than the runner-up and 35 times faster than GPT-6 Sol.
The model, built by post-training Qwen3.5-9B and available through Microsoft Foundry and OpenRouter, can score multiple decisions in parallel and may later be rebased on MAI and OpenAI models.
Analysts said specialized decision models could lower agentic AI workflow costs, but warned that probability scores lack reasoning, can shift with prompt wording and need calibration, audit logs and human review for high-impact uses.