Microsoft-Decision-1: $0.042 per 1M Input Tokens, Under Half OpenAI's Decisions API
- What happened
- Microsoft released Microsoft-Decision-1, a decision-scoring model on Foundry and OpenRouter, at $0.042 per 1M input tokens with free output.
- Why it matters
- It lists at 42% of the $0.10 per 1M input that OpenAI charges for its generally available Decisions API, for yes/no, multiple-choice, rating and rubric grading calls.
- What to do
- If you run classification or LLM-as-judge calls on a chat model, benchmark Decision-1 on your own labelled set before switching.
Decision models are now a price fight, and Microsoft just listed under half of OpenAI's rate. Microsoft-Decision-1 is worth a test for routing, grading and classification calls, measured against OpenAI's Decisions API on GPT-6 Luna. Do not switch on Microsoft's benchmarks alone: every number below is Microsoft's own, and we have not run the model.
What happened
Achint Srivastava, VP of Software Engineering in Microsoft's Office of the CTO, announced Microsoft-Decision-1 on 9 October 2026.
- What it does: "When given a fixed set of answer options, Microsoft-Decision-1 provides a calibrated probability score for each option." It "supports yes/no, multiple-choice, and rating options, as well as rubric-based grading of AI responses and agent actions, all through a simple structured API call." Microsoft positions it for routing, classification, prioritization, verification and workflow control.
- What it is built on: Microsoft says it "post trained Qwen3.5-9B for fast, single-pass decision scoring", and plans to rebase the model on Microsoft AI (MAI) and OpenAI models.
- Where it runs: Microsoft Foundry and OpenRouter.
- Price: "Input tokens cost $0.042 USD per million tokens. Output tokens are free."
- Microsoft's claims: the highest accuracy in a 36-benchmark comparison spanning nearly 150,000 questions kept blind from training, and P50 latency about 35 times faster than GPT-6 Sol and 2.5 times faster than H2O-Lightning-4B v1.1.
Why it matters
| Microsoft-Decision-1 | OpenAI Decisions API (gpt-6-luna) | |
|---|---|---|
| Input, per 1M tokens | $0.042 | $0.10 uncached |
| Output | Free | Not charged |
| Caching | No terms stated | Cache hits discounted, cache writes charged, at standard gpt-6-luna caching rates |
| Extra multipliers | No terms stated | Regional processing premiums and long-context multipliers apply |
On list input price alone, Microsoft's rate is 42% of OpenAI's. OpenAI's guide now marks the Decisions API generally available and says it returns typed answers about 10 times faster than the Responses API, so the comparison is between two shipping products.
The accuracy and latency claims are harder to weigh. Microsoft chose the 36 benchmarks and the comparison set, and published no independent replication. A 9B model beating 26B and 31B models on scoring tasks is plausible, but it is a vendor claim until someone else measures it.
What changes for you
- If you run classification or LLM-as-judge calls on a chat model: you pay only for input, at $0.042 per 1M tokens. Benchmark Decision-1 on your own labelled set before you move traffic.
- If you already use OpenAI's Decisions API: run both on the same set and compare accuracy and cost per call, including OpenAI's cache and regional charges.
- If licensing matters to you: Microsoft's post makes no licence or open-weights statement and gives no context length, even though the base model is Qwen3.5-9B. Ask before you plan self-hosting.
Microsoft-Decision-1 is not in our directory yet.
What to do
- 1 Run Microsoft-Decision-1 and your current classifier or judge model on the same labelled set, and compare accuracy and cost per call before you switch.
- 2 If you plan to self-host, ask Microsoft for licence and open-weights terms first: the launch post states none.
Affected tools & models
Never need to catch up again
The weekly delta: only verdict changes and act-now items. No digest filler.