[KongchangAI]
ModelDecision-1 / Microsoft-Decision-1

Microsoft Decision-1

A specialized decision-scoring model released by Microsoft, post-trained on Qwen3.5-9B. It is designed for scenarios including routing, classification, prioritization, outcome verification, and workflow control, and produces structured probabilistic scores rather than free-form text responses.

Core Facts

Timeline (last 90 days)

Oct 11

Microsoft Decision-1 ranked 6th among 14 cloud-based decision models, answering 75/82 questions correctly, with 28/34 on difficult questions.

Unverified50%
Oct 11

The total cost for the entire cloud testing round was only $0.0195, with Decision-1 costing approximately $0.0036 per 1,000 calls, roughly one seventy-third the cost of calling GPT-6 Sol directly.

Unverified50%
Oct 11

Decision-1 was the only cloud model in the test that was fooled by a prompt injection command hidden in a comment, misclassifying the customer as happy with a probability of 0.68.

Unverified50%
Oct 11

The evaluator recommends that if Decision-1 is used on user-authored text, inputs must be wrapped or sanitized beforehand to prevent prompt injection.

Unverified50%
Oct 10

Microsoft claims Decision-1 is approximately 35 times faster than GPT-6 Sol in median latency.

Unverified50%
Oct 10

Decision-1 is live on Microsoft Foundry, priced at $0.042 per million input tokens, with output being free.

Partially Verified65%

All Facts (6)

Source Articles