SI Agent News
DevelopingClaudeOpenAIResearch

Microsoft launches MAI Code 2 models to compete with OpenAI and Anthropic

The new proprietary models aim to lower costs for developers while reducing Microsoft's reliance on external AI providers.

$0.05Promotional price per million input tokens for MAI Code 2 Flash

The short version

Why it matters

For developers and companies, this means more affordable options for automated coding, but it also signals Microsoft's push to own the underlying technology rather than just renting it from partners. It forces a shift toward evaluating AI based on total software delivery costs rather than just the price of generating lines of code.

A shift toward independence

Microsoft recently introduced a pair of proprietary AI models, MAI Code 2 and MAI Code 2 Flash, during the Ignite 2026 event. These models are specifically designed to handle coding tasks, ranging from complex enterprise debugging to high-volume routine fixes. By building these models in-house, Microsoft aims to establish its own AI capabilities and decrease its current reliance on external partners like OpenAI and Anthropic.

The move is part of a broader strategy to protect profit margins within the developer ecosystem. Because Microsoft owns platforms like GitHub, Visual Studio, and Azure, it can integrate these new models directly into tools used by millions of people. This allows the company to avoid paying licensing fees to external providers every time a user interacts with GitHub Copilot.

Industry analysts suggest that this strategy is about locking developers into a unified ecosystem. By providing its own models, Microsoft can offer a more integrated experience while keeping more of the revenue generated from AI services. The company is positioning itself to be the primary infrastructure provider for the next generation of software development.

Model specializations and benchmarks

The two new models serve different roles within a development workflow. MAI Code 2 is built for heavy-duty tasks that require reasoning across a whole repository, such as implementing changes across multiple files or developing user interfaces. It is intentionally not tuned for extreme problems like fundamental science research or long-horizon agentic work, focusing instead on the daily needs of professional engineers.

In contrast, MAI Code 2 Flash is optimized for speed and high-volume output. It is intended for tasks like explaining existing code, performing targeted edits, and making rapid iterations. Microsoft tested these models against competitors using four common benchmarks, including DeepSWE 1.1 and Terminal-Bench 2.1, to demonstrate their performance in real-world coding scenarios.

The benchmark results show a mixed landscape of performance. While MAI Code 2 recorded an average pass rate of 63.2%—slightly trailing behind Anthropic’s Claude Opus 5.5—the Flash model showed strength in its category. MAI Code 2 Flash achieved an average pass rate of 61% across three benchmarks, which actually outperformed both GPT-6 Luna and Claude Haiku 5.5 in those specific tests.

Promotional pricing and costs

To encourage adoption, Microsoft is offering significant promotional pricing for these models through May 17, 2027. For MAI Code 2, the promotional rate is $0.50 per million input tokens and $2.50 per million output tokens. After the promotion ends, these prices are scheduled to double to $1.00 and $5.00 per million tokens, respectively.

The Flash model is even more affordable during the promotional period, priced at $0.05 per million input tokens and $0.25 per million output tokens. Microsoft used these figures to claim that their models are significantly cheaper than competing offerings from OpenAI and Anthropic for common coding tasks.

However, some analysts caution that these comparisons may be slightly misleading. They point out that the costs Microsoft highlights are based on benchmark runs rather than standard published prices. Furthermore, the gap between Microsoft's costs and its competitors' costs will narrow significantly once the 50% promotional discount expires.

The hidden costs of AI coding

While the raw cost of generating code is decreasing, experts warn that model pricing is only one part of the total expense for a business. Companies must still account for the costs of integration, security testing, and human reviews to ensure the AI-generated code is safe and functional. If model usage only accounts for 10% of a team's total development costs, a 60% reduction in those costs would only result in a 6% overall saving.

There are also concerns regarding the stability and security of AI-generated software. Some research has associated wider AI adoption with weaker delivery stability in software projects. Additionally, Microsoft's own model cards rate both new models highly on offensive cyber capabilities, which could pose risks if not properly managed.

Technology leaders are being advised to look beyond the price per token. Instead, they should track metrics like the cost per merged code change and the frequency of security findings. The goal for enterprises is to reduce the total cost of delivering working software, rather than simply finding the cheapest way to generate lines of code.

What to watch

Try it

Check the GitHub pricing page to see the specific promotional rates for MAI Code 2 and MAI Code 2 Flash.

Sources and links

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by infoworld.com · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

More AI news