Microsoft Rolls Out MAI Code 2 and Flash, Claims 50x Cost Edge
Microsoft launched MAI Code 2 and MAI Code 2 Flash on GitHub Copilot, with the Flash version heading to a limited Microsoft Foundry preview. The company says Flash beat GPT-6.1 in AT&T evaluations while costing 50x less, though independent benchmarks have not confirmed any of it.
- 2 sources report it: cryptobriefing.com, unite.ai.
- The source published it on Nov 17, 2026 at 6:34 PM.
Nothing open. Every check passed.
Microsoft pushed out the third generation of its homegrown coding models on November 17, 2026. MAI Code 2 and MAI Code 2 Flash are both landing on GitHub Copilot, and the Flash variant is also tucked into a limited-access preview inside Microsoft Foundry. That preview means Microsoft is keeping a tight lid on who gets to test it first.
The loudest claim comes out of testing at AT&T. In those evaluations, MAI Code 2 Flash was reported to outperform GPT-6.1, with the research summary narrowing that to certain GPT-6.1 variants rather than the entire family. The same tests put Flash at 50x cheaper than the competition.
There is a catch. Independent benchmarks have not verified the performance claims, and Microsoft has not confirmed what wider availability beyond the Foundry preview looks like. The limited preview also means there is no broad pool of real-world user data yet.
This is a short family history. MAI Code 1 Flash arrived in June 2026 as the first coding model Microsoft built in-house for Copilot. MAI Code 1.1 Flash followed in August 2026, adding native vision support so the model could handle images alongside text, with pricing roughly 73% lower. Those earlier releases reportedly posted higher code accept rates — a measure of how often developers keep the suggestions they are shown — and lower token usage.
The research frames all of this as a step away from leaning on third-party architectures from OpenAI and Anthropic. Owning the model means owning the cost structure, and that gives Microsoft more say over what Copilot users end up paying. For enterprises, the AT&T numbers are the hook: coding bills that scale with headcount and usage are exactly where a 50x cost gap compounds fast.
The risks sit right next to the upside. Vendor-run evaluations are a starting point rather than a verdict, and claims about beating certain GPT-6.1 variants leave room for argument about which variants and which tasks were actually tested.
TopicsMicrosoft · MAI Code 2 · MAI Code 2 Flash · GitHub Copilot · Microsoft Foundry · AT&T · GPT-6.1 · OpenAI
Written by AI from verified reporting.
More stories
- Tech & AI cryptobriefing.com 0 viewsOpenAI ships GPT-6.2 Astra, first deployed model to trip its own self-improvement alarm
- Tech & AI nbcnews.com Caught by @taufik.utami96 1 viewApple Blames Internal Update for Outage Across iCloud, App Store and Store Systems
- Tech & AI livemint.com 1 viewOpenAI ships GPT-6.2 Astra after scrapping the previous model
- Tech & AI breakingthenews.net 0 viewsMicrosoft Debuts MAI Cyber 2 Flash and Perception Cyber Stack at Ignite

