Microsoft Ships Seven Models; Decision 1.1 Claimed 50x Faster Than GPT-Sol 6.1
Microsoft's Ignite launch on 17 November shipped seven models on Maia 200. Decision 1.1 is real: 2.23x faster than v1, Brier 0.1604. The 200x claim isn't Microsoft's.
By Amara Osei
4 min read

What's News
- Microsoft announced seven models at Ignite on 17 November, all running on Maia 200 silicon claiming 1.7x more tokens per watt.
- Microsoft-Decision-1.1 is claimed 50x faster and 79x cheaper than GPT-Sol 6.1; the viral 200x figure appears nowhere in Microsoft's materials.
- Independent testing by Artificial Analysis puts Decision 1.1 at 123.25 ms per task — a 2.23x speedup over version 1 — ranking eighth of eighteen models.
- Decision 1.1 is priced at $0.042 per million input tokens with free output tokens, built on Gemma 4 26B-A4B.
- MAI Code 2's 40% cost claim depends on a 50% promo price ending 17 May 2027; at list price it costs about 81% of GPT-6.1 Sol per task.
Microsoft AI announced seven new models at its Ignite conference on 17 November, and its central claim about the decision model — 50x faster and 79x cheaper than OpenAI's GPT-Sol 6.1 — comes with fine print that most viral posts skipped. The "10x faster and 200x cheaper than a general model" figure that circulated on 17 November, traced to a TestingCatalog post on X, appears nowhere in Microsoft's own materials.
What shipped?
All seven models run in production on Microsoft's second-generation Maia 200 accelerator, which the company says delivers 1.7x more tokens per watt. The lineup:
- MAI Cyber 2 Flash — security; 92.3% on CyberGym-E2E vs 91.5% for GPT-6 Astra, at under 2% of the cost per task at list price ($0.10/$0.50 per million tokens vs Astra's $10/$50)
- MAI Code 2 — enterprise coding; 98% of GPT-6.1 Sol performance at 40% of cost (promo price), and 96% of Claude Opus 5.5 at 5% of the cost
- MAI Code 2 Flash — lightweight coding; claimed to match GPT-6 Luna at 65% of its price
- MAI Image 2.7 — ranked #1 on Artificial Analysis' image-editing leaderboard at roughly one-sixth the price of the four models behind it
- MAI Realtime 1 — speech-to-speech; claimed to beat GPT-Realtime 2.1, starting at $0.01 per audio minute
- Dragon Copilot Transcribe — built on MAI Transcribe 2; powers 10 million patient encounters a month with multilingual error rates halved
- Decision 1.1 — the decision/routing model behind the headline numbers
What is Decision 1.1, exactly?
Microsoft never uses the string "MAI Decision 1." The Ignite post calls it Decision 1.1; the Foundry blog and model card call it Microsoft-Decision-1.1. Version 1 reached general availability on Foundry on 8 October 2026, built on Alibaba's Qwen3.5-9B with no published benchmarks. Version 1.1 switches to Google's Gemma 4 26B-A4B — a mixture-of-experts base with roughly 4B active parameters — adds multimodal input, and ships with a full performance story.
The contract is unchanged: you send a state plus typed questions with fixed option sets, and get probabilities back, not prose. It runs on a custom inference engine at /providers/microsoft/v1/systemone on Foundry, priced at $0.042 per million input tokens (Global Standard) and $0.05 in EU and APAC zones. Output tokens are free. There is no download or fine-tuning path.
Do the speed claims hold up?
Microsoft publishes two scoped comparisons. The first: 50x faster and 79x cheaper than GPT-Sol 6.1 on a single routing task, measured on Microsoft's own harness. The second, per the Foundry launch post: Decision 1.1 is "more than twice as fast as our first model, at the same cost," with a Copilot routing example running 2.2x faster and 4.4x cheaper than a large language model at equal accuracy.
The second claim survives third-party checking. Artificial Analysis measures Decision 1.1 at 123.25 ms per task on its Decision Index v2.1 against 274.25 ms for version 1 — a 2.23x speedup. On cost, AA computes $139.64 per million tasks vs $142.56, about 2% cheaper, because the versions use different token counts per task. The 50x/79x figure is a vendor measurement; the gap against a general model doing the job in JSON mode is real, but the multiple is Microsoft's choice of baseline.
Where does it rank independently?
Decision 1.1 scores 67.78 on the Decision Index v2.1 (read 18 November 2026), eighth of eighteen models. Microsoft's launch post says it scores 68, up six points, ahead of Jev 1.14 (66.23), Gemini Decision 1 (65.98) and GPT-6 Luna Decisions (60.25) — all true. What the post omits: seven models score higher, and leader SubQ Snap 1.2 sits 7.1 points clear. The most meaningful number is calibration — the Brier score improved from 0.1965 to 0.1604, meaning the probabilities got closer to true. Accuracy rose from 40.59 to 48.45.
The pricing fine print
MAI Code 2's "40% of the cost" claim rests on a 50% promo price ending 17 May 2027. At list price ($1/$5 per million tokens vs promo $0.50/$2.50), it costs about 81% of GPT-6.1 Sol per task. MAI Code 2 Flash is 30% more expensive per task than GPT-6 Luna at list price, though it scores higher. MAI Cyber 2 Flash's cost claim holds at list price, but Microsoft excluded 92 of 905 CyberGym-E2E tasks (10.2%) that overlapped its training data, and the model ships only inside Microsoft security products (MDASH from mid-December).
Microsoft also announced MAI Licensed Weights — custom MAI models in customers' own Azure environments, coming via Frontier Company engagements. Decision 1.1 is already GA on Foundry, OpenRouter and Vercel, with 4,000+ customers per Microsoft. The model is real; the 200x ratio in the tweet is not.
Source: epyhia.com
More from Amara Osei
Show full bio
Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.
2,248 articles