Microsoft's MAI Models Bet on Price: Code 2 at 40% of Sol, Cyber 2 Flash Under 2% of Astra
Microsoft AI released a family of specialised MAI models at Ignite on November 17, competing on price rather than peak capability. It says MAI Code 2 reaches 98% of GPT-6.1 Sol's coding performance at about 40% of the price and 96% of Claude Opus 5.5's at 5%, and that MAI Cyber 2 Flash beats GPT-6 Astra on one security benchmark at under 2% of the cost per task. It also offered enterprises licensed MAI weights to customise and run in their own Azure environments. GitHub said a pilot using Microsoft's MDASH scanner produced 99 checked vulnerability reports for six open-source projects, 55 of them fixed. The benchmark comparisons are Microsoft's own.

Microsoft is not yet claiming the frontier; it is claiming the bill. The models Microsoft AI released at Ignite on November 17 are specialised, and each is sold against a named rival on price: most of the quality of an OpenAI or Anthropic model for a fraction of the cost. The pitch matters because Microsoft can route its own traffic, starting with GitHub Copilot, to whichever model it chooses.
The lineup#
| Model | Microsoft's claim |
|---|---|
| MAI Code 2 | 98% of GPT-6.1 Sol's coding performance at about 40% of the price; 96% of Claude Opus 5.5's at 5% |
| MAI Code 2 Flash | Matches GPT-6 Luna at 65% of the price and Claude Haiku 5.5 at 20% |
| MAI Cyber 2 Flash | Beats GPT-6 Astra on CyberGymE2E at less than 2% of the cost per task |
| MAI Image 2.7 | First on Artificial Analysis image editing |
| MAI Realtime 1 | Speech-to-speech model from $0.01 per audio minute |
| Decision 1.1 | Routing model Microsoft says is 50 times faster and 79 times cheaper than GPT-6.1 Sol |
Microsoft says the coding figures average four benchmarks drawn from everyday enterprise work (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1), and that the models are "explicitly not yet tuned" for harder problems such as fundamental science research and very long agentic work. MAI Code 2 is available in GitHub Copilot on Pro, Pro+, Max, Business and Enterprise plans, and Code 2 Flash on all plans, GitHub said. Microsoft also says its Maia 200 accelerator is now in production running its largest models.
Two strategic pieces sit outside the table. MAI Licensed Weights lets enterprises take a MAI model, customise it with their own data and run it in their own Azure environment, which Microsoft AI chief Mustafa Suleyman told Axios would "feel like open source" with more security and control. And Suleyman said Microsoft is still building a general-purpose frontier model, with state-of-the-art pre-training results, but gave no release timeline: "My team is building true superintelligence."
The cyber angle, with a check#
MAI Cyber 2 Flash is aimed at defenders running scanners around the clock. Microsoft says MAI Cyber models already account for almost 40% of scans in MDASH, the agentic scanning harness it introduced in May, and that the new model's output tokens cost 86% less than MAI Cyber 1.1 Flash's.
GitHub published what that pipeline looks like with humans in the loop. In a pilot with Microsoft, MDASH running the earlier MAI Cyber model flagged thousands of possible issues across six open-source projects: AutoGPT, Caddy, Express, FastMCP, ImageMagick and Netty. After AI triage and hand review by seven Microsoft researchers, maintainers received 99 private reports with suggested fixes; by November 15 they had fixed 55, and 14 of those fixes had shipped in a release. GitHub's own caveat is the useful one: the researchers spent about two days per project checking findings by hand, and that work "doesn't scale yet." It is the defensive counterpart to the automated exploit pipelines Google says attackers are already running.
Limits#
- Every performance and price comparison in the table is Microsoft's, on benchmarks it selected; we found no independent replication. Axios notes that some of the pricing advantages reflect promotional discounts.
- Microsoft's post counts seven new models; Axios counted six. We list the models Microsoft names.
- The MDASH pilot covers six projects that agreed to take part, which GitHub calls a small sample.
Sources#
- Microsoft AI: Seven new world class models at scalable price-performance (November 17, 2026)
- Axios: Microsoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI models (November 17, 2026)
- GitHub Changelog: MAI Code 2 and MAI Code 2 Flash in GitHub Copilot (November 17, 2026)
- The GitHub Blog: Maintainers don't need more AI bug reports. They need fixes. (November 17, 2026)