Microsoft launches high-speed decision AI model, 35 times faster than GPT-6 Sol

Microsoft is strengthening its AI product line with a specialized strategy.

On Friday, Microsoft CEO Satya Nadella announced the launch of a new generation fast-decision AI model Microsoft-Decision-1, specifically designed for structured decision-making tasks.

In a post on platform X, Nadella stated that this model "performs excellently on structured decision-making tasks, surpassing large language models and other decision models in terms of both latency and quality," and it has already been tested internally by Microsoft in scenarios such as incident response, quality control, and scientific research.

Microsoft claims that in internal testing, the model runs 35 times faster than OpenAI’s GPT-6 Sol and 4.5 times faster than Quyet-1.0-Large.

From a pricing perspective, the model charges $0.042 per 1 million input tokens, with output being free, making this arrangement particularly attractive cost-wise for high-frequency, repetitive decision-making applications.

Unlike traditional large language models that focus on text generation and complex reasoning, the design goal of Microsoft-Decision-1 is more targeted.

The Microsoft-Decision-1 model selects the optimal solution from preset options and assigns probability scores to each alternative, helping downstream applications to determine whether to proceed, retry, escalate, or hand off to manual review.

According to Microsoft, the model achieved top accuracy across 36 benchmarks covering nearly 150,000 questions. Its functional positioning includes routing, content classification, task prioritization, result verification, and workflow control, and it can be seamlessly embedded in existing applications, AI agents, and workflow systems.

However, these performance figures are based on Microsoft’s own benchmarks—independent third-party verification is not yet complete.

Multiple internal Microsoft teams have conducted real-world scenario testing of the model.

The Xbox Research team used it to classify over 10,000 pieces of game feedback, and results showed quality comparable to GPT-6 Sol, but with more than 14 times the speed and about 200 times lower cost. The Microsoft Copilot team also found the model performed similarly to GPT-5.6 Luna on AI response evaluation tasks.

Technically, Microsoft-Decision-1 is built on Qwen3.5-9B and has undergone targeted post-training focused on single-decision scoring.

Microsoft also plans to migrate the model in the future to other foundations, including its MAI series models and OpenAI models.

In the announcement, Microsoft emphasized that the model is designed “to embed decision intelligence into existing applications in a safe and trustworthy environment,” establishing it as one of the foundational components for AI agent workflow orchestration.

Currently, Microsoft-Decision-1 is available to developers through Microsoft’s AI development platform, Microsoft Foundry, and support for integration via OpenRouter is also in the works.