Microsoft is developing its own AI models to handle more of the work inside GitHub Copilot. At Ignite on November 17, it introduced two coding models built for different kinds of work, including a lower-cost option for routine edits and code explanations.
The details
- Code 2 handles demanding tasks such as debugging and changes across multiple files. Code 2 Flash is designed for faster, higher-volume work like targeted edits and routine fixes.
- Both models are generally available in GitHub Copilot. Code 2 Flash is also in limited-access preview in Microsoft Foundry. Microsoft separately introduced Cyber 2 Flash for security work and licensed access to its model weights for select customers.
- Microsoft says Flash delivers strong performance on everyday coding tasks at a lower cost than competing models. Those comparisons are based on Microsoft's evaluations and promotional pricing.
The business angle
If Flash can handle routine work at the quality users expect, Microsoft can reduce the cost of serving Copilot requests and reserve more expensive models for harder jobs.
GitHub gives Microsoft an established channel for distributing its own models to developers. Foundry offers a potential second business: selling those models to customers building AI applications of their own.
Get Boulton in your inbox
Every Thursday

