MAI Code 2 and Code 2 Flash
Points and comments are a snapshot, not live.
Microsoft's MAI Code 2 models promise frontier-level coding performance at 5-40% of competitors' costs.
Microsoft AI introduced MAI Code 2 and MAI Code 2 Flash, coding models for enterprise tasks like debugging, multi-file implementation, and repo-wide reasoning.
Averaged over four benchmarks (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, FrontierCode v1.1), MAI Code 2 achieves 98% of GPT 6.1 Sol's performance at 40% of the cost and 96% of Claude Opus 5.5's at 5% of the cost. MAI Code 2 Flash offers comparable performance to GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of the cost, respectively. In blind frontend app evaluations, reviewers preferred MAI Code 2 over GPT-6 Sol 64% of the time.
What commenters are saying
Commenters were skeptical of Microsoft's benchmark comparisons, noting they may compare against premium pricing tiers or promotional discounts rather than standard rates. A camp questioned whether Microsoft could persuade enterprises to switch from Anthropic or OpenAI models, while others countered that cost reductions are positive for consumers. Several users expressed interest in trying the models for routine coding tasks currently handled by Haiku or Luna. A technical discussion arose about the models' foundations, with one user linking to Microsoft's Frontier Pre-Training announcement. The thread also debated font pairing choices on the announcement page and whether Microsoft was backing away from the Copilot branding.