MAI Code 2 and Code 2 Flash
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
Microsoft released MAI Code 2 and Code 2 Flash, now in GitHub Copilot under usage-based billing.
Microsoft claims near-rival quality at a fraction of the price; third-party checks pending.
Background
Microsoft AI (MAI) is Microsoft's AI unit, and GitHub Copilot is GitHub's developer product where the models appear in the model picker. The MAI Code 2 family is described as Microsoft's agentic coding models, extending the earlier MAI-Code line from lightweight coding assistance to complex software engineering, built on new larger pre-trained bases with a 1M-token context window and configurable reasoning effort. The release came among seven new models at Microsoft Ignite 2026, roughly three months after the first MAI-Code-1-Flash debuted at Build in June 2026.
How it unfolded
- Jun 2, 2026Microsoft announced MAI-Code-1-Flash, its first coding model, at its Build developer conference in San Francisco, and it began rolling out in GitHub Copilot.
- Aug 11, 2026MAI-Code-1.1-Flash, adding native vision support, became available in GitHub Copilot at a 73% lower list price than MAI-Code-1-Flash.
- Aug 2026 to Nov 2026MAI Code 2 and MAI Code 2 Flash were trained.
- Nov 17, 2026Microsoft released MAI Code 2 and MAI Code 2 Flash, with both available in GitHub Copilot under usage-based billing and Code 2 Flash also in a limited access preview in Microsoft Foundry.
Key terms
- GitHub Copilot
- An AI coding assistant that autocompletes code inside popular integrated development environments.
- usage-based billing
- A limit set by an internet provider on how much data a customer can transfer, with extra fees beyond it.
- context window
- The maximum number of tokens a large language model can consider at one time when generating an output.
Source: Wikipedia
Who’s saying what
- Official
- Microsoft Developer says Code 2 brings frontier price-performance for everyday enterprise coding tasks, while Code 2 Flash offers affordable, high-quality coding for routine work at scale.
- Official
- Microsoft AI says MAI Code 2 reaches 98% of GPT-6.1 Sol's performance at 40% of the cost and 96% of Claude Opus 5.5's at 5% of the cost, and that developers preferred it over GPT-6 Sol 68% of the time in a blind frontend evaluation.
- Caution
- X users questioned the claims, with one saying that if the models were any good Microsoft would publish them to Artificial Analysis, and another asking why they are not on OpenRouter and why people are stopped from using them outside Microsoft products.
Still unverified
Weizhu Chen said more third-party evaluations will be shared soon to demonstrate the models' capabilities; those results have not yet been published.