Microsoft releases MAI Code 2 with discounted launch rates
Microsoft’s comparisons with OpenAI and Anthropic models draw on four coding benchmarks it selected and a temporary 50% pricing discount.

Microsoft unveiled its MAI Code 2 coding model at Ignite in San Francisco, saying it offers 98% of the performance of OpenAI’s GPT-6.1 Sol for 40% of the price. GitHub Copilot users can access the model and its smaller counterpart, MAI Code 2 Flash.
The company’s comparison with Anthropic puts Code 2 at 96% of Claude Opus 5.5’s performance for 5% of its cost. Microsoft based the figures on average results across four coding benchmarks of its choosing.
The pricing claims incorporate a 50% introductory discount disclosed in a Microsoft AI announcement footnote. GitHub lists the offer as running until 17 May 2027. Once it ends, the price for a million input tokens will increase from US$0.50 to US$1; a million output tokens will cost US$5 instead of US$2.50.
Microsoft puts Flash’s performance on a par with GPT-6 Luna for 65% of its price, and with Claude Haiku 5.5 for 20%. The company cautions that neither coding model is yet tuned for tasks such as fundamental science research or extended agentic work.
A separate offering, MAI Licensed Weights, will let trusted customers adapt Microsoft AI models with their own data and run them in their own Azure environments. Microsoft’s Frontier Company team will assist them.
Microsoft AI chief executive Mustafa Suleyman told Axios that work is also underway on a general-purpose frontier model. He did not specify when it would be released.




