MAI Code 2 addresses complex production tasks, visual coding, and repository-wide reasoning, while the Flash variant focuses on rapid, high-volume iterations like targeted edits. Benchmarks show the primary model matches up to 98% of rival frontier performance at as little as 5% to 40% of the cost. Beyond standalone coding environments, an optimized variant named MAI Work Flash is being integrated into Microsoft 365 and Fabric IQ to handle business queries and office automation. Both core models are now available through GitHub Copilot, with broader distribution planned for third-party developer platforms.
Key points
The new models deliver enterprise coding and debugging capabilities at significantly redu…: The new models deliver enterprise coding and debugging capabilities at significantly reduced operating costs.
Neither model is currently tuned for long-horizon agentic work or fundamental scientific…: Neither model is currently tuned for long-horizon agentic work or fundamental scientific research.
Access is currently live in GitHub Copilot, with future rollouts scheduled across externa…: Access is currently live in GitHub Copilot, with future rollouts scheduled across external platforms like Vercel and Open Router.
Sources checked by our analysis pipeline: 1
MAI Cyber 2 Flash: Leading security. Lower cost
MAI Code 2 and MAI Code 2 Flash: two new coding models built for the everyday enterprise coding tasks, delivering leading performance at a fraction of the cost of competing models.
MAI Code 2 is at the frontier of price-performance for common enterprise coding tasks and excels in taking code to production, deep debugging, multi-file implementation, repo wide reasoning, and UI. It does all of this while also delivering changes code maintainers would merge, at a fraction of frontier cost.
Meanwhile, MAI Code 2 Flash is designed for high-volume work such as code explanations, targeted edits, routine fixes, and rapid iterations, while bringing quality at a much lower cost.
To deliver the best price-performance, we’ve built these models to prioritize the coding work our customers do every day. They’re explicitly not yet tuned for harder problems, like fundamental science research and very long horizon agentic work.
Averaged over four widely used benchmarks that target the everyday enterprise coding tasks we see in GitHub Copilot (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode…
The search giant revised its data management terms to hold onto sensitive interaction logs significantly longer than previously permitted. Users relying on history-off settings or ephemeral sessions will now have their conversational records preserved by the platform for several weeks. This shift marks a notable expansion in how long opt-out interactions remain stored on company infrastructure.
The search giant is expanding consumer access to autonomous digital assistance by dropping the subscription paywall for its flagship tool. Driven by the new Gemini 4 Hydrogen model, the system handles everyday personal logistics like calendar scheduling, meal planning, and expense tracking directly within Gemini. This broader release intensifies industry competition against rival offerings from OpenAI, Meta, and Anthropic.
The retention policy update applies globally to the Gemini app over the coming weeks to strengthen safety protections and address technical system failures. Concurrently, new privacy controls will separate user conversation storage from data contribution pipelines. This change gives individuals granular management over whether their logged interactions can be utilized to train future artificial intelligence models.