HOT RADAR
llmHEAT 25.1°

AI CURATED HOT STORY · 11/18/2026

Microsoft Unveils MAI Code 2 with 5% of Rival's Cost

2 reports archived2 independent sourcesupdated 11/18/2026, 06:42:53
Synthesis & Latest Updates
2 Sources Cross-Validated

Microsoft has officially launched its self-developed coding models MAI Code 2 and MAI Code 2 Flash at Ignite 2026, now available on GitHub Copilot. Official data shows that MAI Code 2 achieves about 98% of GPT-6.1 Sol's performance on four coding benchmarks, with a per-task cost of only $0.309—approximately 5% of Claude Opus 5.5's cost—while delivering about 96% of the latter's performance. The Flash version achieves an average pass rate of 61.0% on three tests including DeepSWE 1.1, surpassing GPT-6 Luna and Claude Haiku 5.5.

Both models support tool calling and step-by-step task execution, covering enterprise scenarios such as code production, deep debugging, multi-file implementation, repository-level reasoning, and UI generation. MAI Code 2 focuses on price-performance leadership for everyday enterprise coding, while Flash further reduces cost while offering similar capabilities. Microsoft enters the code generation race with its own models, competing directly with mainstream coding models.

This launch marks Microsoft's shift from relying on external models to self-developed core models in AI coding tools, potentially significantly reducing costs for enterprise developers and intensifying price competition in the coding model market. GitHub Copilot subscribers will directly benefit from lower inference costs and more cost-effective options.

LATEST/Microsoft launched MAI Code 2 and Flash at Ignite 2026, now integrated into GitHub Copilot. The company emphasizes price-performance leadership, with MAI Code 2 costing only 5% of Claude Opus 5.5 while delivering 96% of its performance, and Flash outperforming competitors on multiple benchmarks. This marks Microsoft's formal entry into the coding model market with self-developed models.

Hourly heat curve

7 fully observed hours
Now
25.1
Peak
26.207:00
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 2 reports · Latest first