GitHub Copilot users can now choose Microsoft-built AI models instead of routing every request through OpenAI or Anthropic. At Build 2026 on June 2, Microsoft launched MAI-Code-1-Flash (a coding model) and MAI-Thinking-1 (a reasoning model) — the company’s first models built entirely in-house. The timing matters: GitHub Copilot switched to token-based billing on June 1. Which model your team uses now has a direct line to what you pay.

Key takeaways:

• MAI-Code-1-Flash is live in the GitHub Copilot model picker (VS Code); MAI-Thinking-1 is in private preview on Azure Foundry.
• Microsoft’s own benchmarks: 60% fewer tokens vs. Claude Haiku 4.5 on coding tasks — independent confirmation not yet available.
• Indicative pricing: $0.75/M input and $4.50/M output tokens for MAI-Code-1-Flash — marked “being finalized” by GitHub.
• Microsoft’s AI chief claimed 10× cost efficiency vs. GPT-5.5 for McKinsey workloads — vendor-reported, not independently benchmarked.

What Did Microsoft Launch at Build 2026?

MAI-Code-1-Flash is Microsoft’s first in-house coding model. Trained on GitHub Copilot’s production harnesses rather than benchmark leaderboards, it scored 51.2% vs. 35.2% for Claude Haiku 4.5 on SWE-Bench Pro — a 16-point lead — while completing tasks using up to 60% fewer tokens. It is rolling out to Copilot individual users in VS Code’s model picker starting today.

MAI-Thinking-1 is Microsoft’s first in-house reasoning model, targeting enterprise architecture, compliance, and complex decision workflows. Per CNBC’s Build 2026 coverage, AI CEO Mustafa Suleiman said it outperformed GPT-5.5 at 10× better cost efficiency for McKinsey’s workloads — a vendor-reported, workload-specific claim with no independent benchmarks yet. MAI-Thinking-1 is in private preview through Microsoft Foundry.

Why Does Model Choice in Copilot Matter Right Now?

Model selection in Copilot was mostly a quality preference until June 1, when GitHub switched to consumption-based billing. Agentic coding tasks — code review, multi-file refactoring, test generation — consume thousands of tokens per session. Defaulting to GPT-5.5 for everything means premium pricing on every token. MAI-Code-1-Flash, if its efficiency claims hold in your workflows, offers a substantially lower rate for the same task category.

This is also Microsoft’s structural move toward pricing independence from OpenAI — a shift that accelerated after the Microsoft-OpenAI relationship changed materially in April 2026. Running MAI models on Azure avoids OpenAI royalties, a cost advantage Microsoft can pass to enterprise customers.

Near-term move: Evaluate MAI-Code-1-Flash on routine tasks — inline completion, refactoring, documentation — where a cheaper, efficient model reduces consumption cost with minimal risk. Reserve frontier models for complex or high-stakes work. Ask your GitHub or Azure account team which MAI models are available under your enterprise agreement, how costs are metered, and when MAI-Thinking-1 exits preview.

---

FAQ

Does switching to MAI-Code-1-Flash break my existing Copilot workflows? No. MAI-Code-1-Flash appears as an additional choice in Copilot’s model picker. Your current model defaults remain; you opt in via the picker or auto-picker. Switching is reversible.

How does MAI-Code-1-Flash pricing compare to GPT-5.5? Published pricing is $0.75 per million input tokens and $4.50 per million output tokens — though GitHub notes this is still being finalized. GPT-5.5 pricing through GitHub Copilot is higher. Independent price comparisons have not yet been published; verify current rates directly with your GitHub enterprise agreement before making budget commitments.

Should enterprises use MAI-Thinking-1 for compliance or legal workflows? Too early. MAI-Thinking-1 is in private preview with no independent benchmarks and no disclosed pricing. Microsoft’s onstage efficiency claim (10× better cost vs. GPT-5.5 for McKinsey workloads) is vendor-reported and workload-specific. Express interest in the preview, test it on non-sensitive tasks first, and wait for peer-reviewed or customer-published comparisons before relying on it for regulated workflows.