MAI Code 2 Flash

MAI Code 2 Flash is the lighter, high-volume coding model in Microsoft AI’s MAI coding line. Announced at Microsoft Ignite on 2026-11-17 alongside mai-code-2. Part of the MAI series.

Positioning (Microsoft’s post)

  • For high-volume work: code explanations, targeted edits, routine fixes and rapid iterations.
  • Like Code 2, it is not tuned for fundamental science research or very long-horizon agentic work.

Microsoft’s claims

  • Across three day-to-day coding benchmarks (the post does not name them), it performs comparably to GPT-6 Luna and Claude Haiku 5.5, at 65% and 20% of their relative cost.
  • The post’s footnotes give the same test conditions as mai-code-2 (maximum reasoning effort; MSBench with the GitHub Copilot CLI for some benchmarks).

MAI Work Flash

The post also announces MAI Work Flash, a variant optimised from Code 2 Flash for Microsoft 365 and FabricIQ. It has no separate note. Microsoft’s claims for it are:

  • Excel and PowerPoint: on par with GPT-6 Luna on accuracy or quality and token efficiency, on internal benchmarks.
  • Word: similar to GPT-6.1 Sol across five internal benchmarks, using about 42% fewer input tokens per call in production.
  • Teams: beats GPT-6 Luna on tracking decisions, filling knowledge gaps and following custom instructions, with up to 28% fewer tokens.
  • FabricIQ: in offline tests, outperforms GPT-5.6 Sol, GPT-6 Astra and Claude Opus 5.5 at DAX query generation.

Availability

  • GitHub Copilot: available.
  • Microsoft Foundry: limited-access preview.
  • Other channels: OpenRouter, Vercel, Hermes, OpenCode and Kilo are listed as coming soon, with no dates.

Billing

As for mai-code-2, the post gives only relative cost percentages and a footnoted 50% promotional rate, with no absolute prices. Check Microsoft’s pricing page for billing names and tiers.

Source limits

  • No context window, parameter count or model card is stated in the post.
  • The three benchmarks behind the GPT-6 Luna and Haiku 5.5 comparison are not named, so that comparison cannot be checked from the post alone.
  • All figures are Microsoft’s own.

Sources