Breaking

Microsoft Unveils Its Own Coding Model MAI-Code-1 at Build

June 3, 2026 at 13:50 EDT

On June 2, 2026, Microsoft unveiled its in-house coding models "MAI-Code-1" and the smaller "MAI-Code-1-Flash" at its Build 2026 developer conference, optimizing them for GitHub Copilot and beginning rollout. Part of a family of homegrown models aimed at reducing reliance on OpenAI and cutting costs, the originating @MicrosoftAI announcement points to the Build session "Microsoft AI coding models are optimized for GitHub Copilot."

June 2, 2026 · Microsoft Build 2026

Microsoft brings coding AI in-house with MAI-Code-1

Two homegrown models — MAI-Code-1 and the smaller MAI-Code-1-Flash — roll into GitHub Copilot across every plan, trained directly on Copilot's production harness to cut cost and reliance on OpenAI.

~5B

active parameters (137B total) — a compact model

256K

token context length

−60%

tokens used on complex tasks → lower cost

new MAI models in the family

SWE-Bench Pro — MAI-Code-1-Flash vs Claude Haiku 4.5

Microsoft's internal evaluation · higher is better

MAI-Code-1-Flash51.2%

Claude Haiku 4.535.2%

+16 points ahead on SWE-Bench Pro — and 71.6 vs 66.6 on SWE-Bench Verified.

+28.9

points · IF Bench (instruction following)

85.8%

adjusted accuracy · adversarial coding

Why it matters

Trained on Copilot's production harness → Less reliance on OpenAI → Lower cost & faster agentic coding

"Zero distillation" on clean, commercially licensed data · trained March–May 2026 · available now in VS Code Copilot across Free, Pro, Pro+ and Max.

Welcoming view

Microsoft-made, not borrowed from OpenAI
Rolled out to all Copilot tiers
Cheaper and faster — 60% token reduction

Cautious view

Benchmarks are largely Microsoft's internal evals
Awaiting independent verification
Real savings hinge on Copilot's token billing

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.