Overview
MAI Code 2 and MAI Code 2 Flash are Microsoft's in-house coding models, announced at Ignite on November 18, 2026, and live in GitHub Copilot the same day. The split is straightforward: Code 2 handles complex software development, Flash handles high-frequency everyday coding. Microsoft is not selling the strongest coding model. It is selling a tier that is good enough and cheap.
The published numbers come in two comparisons. Across an average of four benchmarks, Code 2 reaches 98% of GPT-6.1 Sol's performance at 40% of the cost, and 96% of Claude Opus 5.5's performance at 5% of the cost. In blind reviews of frontend code, reviewers preferred its output 64% of the time against GPT-6 Sol.
There is a reading problem worth naming. Both 96% and 98% are averages over four benchmarks, and averages flatten the distribution: all four could sit near 96%, or three could be strong while one falls far behind. A team that selects on the average alone can end up on the wrong side of whatever got averaged away. The useful move is to look at what those four benchmarks are and match them to the dominant task type in your codebase.
The pricing structure shows where Microsoft is aiming. A model at 5% of flagship cost can run on every commit, every completion and every CI pass, rather than only when a person asks. Coding is used far more often than most AI tasks, so a low unit price compounds with call volume in a way a low monthly fee never does.
Access is worth being precise about. Both models ship inside GitHub Copilot, riding the existing subscription, so enterprises do not need a separate procurement path. Code 2 takes the work that touches multiple files and spans modules; Flash takes completions, unit tests and small refactors. The deciding question is how many files a change touches, not how impressive the task sounds.
Key Features
- 96% of Opus 5.5 across four benchmarks: Reaches 96% of Claude Opus 5.5's performance on a four-benchmark average at 5% of the cost. An average, not a per-benchmark guarantee.
- 98% of GPT-6.1 Sol at 40% of the cost: On the same four-benchmark average it reaches 98% of GPT-6.1 Sol's performance for 40% of the cost, putting near-peer coding ability at four tenths the price.
- 64% win rate in blind frontend review: In blind reviews of frontend code, reviewers preferred its output 64% of the time against GPT-6 Sol.
- Two tiers split by task: Code 2 for complex software development, Code 2 Flash for high-frequency everyday coding. The split follows task complexity rather than model size.
- Part of the Ignite batch: Announced alongside MAI Image 2.7, MAI Realtime 1 and MAI Cyber 2 Flash, forming the coding line in Microsoft's seven-model MAI release.
- Straight into GitHub Copilot: Live in GitHub Copilot on launch day through the existing subscription, so enterprises skip a separate procurement and integration cycle.
Use Cases
- Everyday enterprise coding: completions, unit tests and small refactors on the Flash tier
- Multi-file changes and cross-module work on Code 2, decided by how many files a change touches
- Frontend development: a 64% blind-review win rate for interface and interaction code
- Automated checks and fixes in CI: 5% of flagship cost makes per-commit runs affordable
- Large-scale changes across private repositories: a low unit price is what makes backlog work viable
- Teams already on GitHub Copilot: pilot without new procurement
Pros
- 96% of Claude Opus 5.5 across four benchmarks at 5% of the cost sets a clear price-to-capability position
- 98% of GPT-6.1 Sol at 40% of the cost makes migration arithmetic straightforward
- A 64% blind-review win rate on frontend code is verifiable evidence for interface work
- Code 2 and Flash split by task complexity instead of defaulting everyone to the priciest tier
- Shipping inside GitHub Copilot on launch day keeps the integration path short
- Shared procurement and deployment with the rest of the MAI batch lowers expansion cost
Pricing
How it is quoted: Microsoft gives relative cost rather than a per-token rate card. Code 2 costs 40% of GPT-6.1 Sol and 5% of Claude Opus 5.5. Performance caveat: those percentages are averages over four benchmarks, not per-benchmark results. Availability: live in GitHub Copilot on launch day as part of the seven-model MAI release at Microsoft Ignite 2026, billed through existing enterprise subscriptions. Sister model: MAI Code 2 Flash targets high-frequency everyday coding as the lightweight tier of the same line.
Summary
MAI Code 2 and Code 2 Flash are Microsoft's in-house coding models, announced at Ignite on November 18, 2026, and live in GitHub Copilot the same day. Code 2 reaches 98% of GPT-6.1 Sol's performance at 40% of the cost and 96% of Claude Opus 5.5's at 5%, measured as a four-benchmark average, and wins 64% of blind frontend reviews against GPT-6 Sol. The two tiers split by task complexity, with Code 2 on complex software development and Flash on high-frequency everyday coding. Microsoft publishes relative cost rather than a per-token rate, and the performance figures are averages rather than per-benchmark results.
Version History
- MAI Code 2 and Code 2 Flash launch (2026-11-18): Microsoft releases two in-house coding models at Ignite 2026, live in GitHub Copilot the same day. A four-benchmark average puts Code 2 at 98% of GPT-6.1 Sol's performance for 40% of the cost and 96% of Claude Opus 5.5's for 5%, with a 64% blind-review win rate against GPT-6 Sol on frontend code.