Microsoft-Decision-1 is live on OpenRouter. Microsoft's numbers: highest accuracy across 36 blind benchmarks (~150K questions), 4.5x faster than the runner-up, 35x faster than GPT-6 Sol, and decisions flip on just 1.3% of perturbed inputs. Post-trained from Qwen3.5-9B. $0.042/M input, output free, 32K context openrouter.ai/microsoft/micr…

Microsoft-Decision-1 - API Pricing & Providers

Microsoft-Decision-1 is a small model built for fast decision-making. $0.042 per million input tokens, $0 per million output tokens.

openrouter.ai
Satya Nadella
@satyanadellaOct 9
Introducing Microsoft-Decision-1, our new model for fast decision-making. It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality. We’re already testing it across Microsoft for everything from incident response and quality control to scientific discovery.

Oct 9, 2026 · 10:43 PM UTC

35
65
14
604
47,707
RelevantRecentLikes
Output free at that accuracy and speed is crazy value
122
people reverse engineer ms products now. an ms did the same. nothing new.
2
781
As a paying customer your service is beginning to suck badly.
84
Guys what’s going to happen to indies if the big ones won’t eve leave the decision model space alone.
299
No fault to openrouter, but i hope no one pays to use Qwen 9b… 🤪
254
$420/btok … trillion is still super expensive for SMBs, $420,000… x1000 price point improvement needed for Excel level innovation for $MSFT !!
76
that flip rate of 1.3 percent sounds incredibly stable
120
@grok how does this compare to jev
1
494
highest accuracy across 36 blind benchmarks is a big claim. which decision types is it optimized for — routing, planning, task decomposition, or something else?
276
wait it's only a 9b qwen post-train? kinda makes you wonder what the big models are wasting on simple yes/no calls
1
226
This new fashion of decision models is sooooo boring… I hope it’s going to be very quick.
72
Incredible performance at an unbeatable price a huge win for developers
7
adding to the benchmark bench.jakecuth.com
33
🌞🫶🏼
239
pricing on that is insane tbh. dropping qwen-based decision models this fast makes automation pipelines so much cheaper to run.
182
it's good to see a chart that shows the full graph instead of part of it. also, 23 of the 36 banchmarks averaging 83.5 latency accuracy for a model to make decisions is good. with a 64 percent passing rate, to me 23 of 36 seem like the model is inconsistent.
188
35x faster than sol and it only flips 1.3% of the time. my agents flip on every tool call
1
441
why are you guys using gpt-6 sol as comparison here? They are totally different models with different purposes
1
2
498
That is why Open Source is so important. This is not an achievement solely of Microsoft, but also of the Qwen developers.
17
Decision-1 is live here too — we tested it in a Snake game against Jev and the difference in decision style is really visible in practice: youtu.be/NqiUcC-AUiI

Microsoft Decision-1 Explained 🔥

Can an AI model make better decisions instead of just generating te...

youtube.com
3
copied jev, atleast are they better than jev?
7
What about the 85ms response time? It's responding in 300ms, more than jev
379
Flip rate on perturbed inputs measures paraphrase, not routing: agents replay the same decision as context grows, so test consistency across context length.
220
Load more replies