MAI-Code-1-Flash

Retired · Language, Code

MAI-Code-1-Flash is a coding model from Microsoft AI, released on 2 June 2026 and offered only inside GitHub Copilot, where GitHub deprecated it across all Copilot experiences on 10 September 2026. A sparse mixture-of-experts model with 5 billion active and 137 billion total parameters, it was trained from a MAI-Thinking-1 checkpoint and with the GitHub Copilot production harness. [1] [2] [3] Primary source

Timeline of MAI-Code-1-Flash →

Claims and evidence

  • Released Gradual rollout to GitHub Copilot individual users in Visual Studio Code, starting with a limited set of users. Primary source[1]dateline; first paragraph [2]Model Summary: Release date [4]
  • Deprecated Date of the deprecation notice. Primary source[5]
  • Retired GitHub calls this date the deprecation date; on it the model was deprecated across all Copilot experiences, its only channel. Primary source[3] [5]
  • Available on more GitHub Copilot surfaces Primary source[6]
  • Generally available for Copilot Business and Enterprise Primary source[7]
  • Status RetiredGitHub deprecated it across all GitHub Copilot experiences on 2026-09-10 and filed the post under Retired; GitHub Copilot was its only distribution channel per the model card. Primary source[3] [2]Distribution channels
  • Replaced by MAI-Code-1.1-FlashSuggested alternative in GitHub Copilot. Primary source[5] [3]
  • Derived from (other) MAI-Thinking-1Training started from MAI-Thinking-1's mid-training checkpoint and continued with supervised fine-tuning, a further mid-training phase on synthetic agentic tasks and reinforcement learning. Primary source[2]Training disclosure; Model dependencies
  • Change · Size Smaller: 5B active and 137B total parameters, against 35B active and about 1T total for MAI-Thinking-1, from whose mid-training checkpoint it was trained.Compared with MAI-Thinking-1 Primary source[2]Model Summary: Parameters; Training disclosure [8]Medium-sized model, with strong software engineering performance
  • Change · Other Specialised for coding: trained further on synthetic agentic tasks and with reinforcement learning and the GitHub Copilot production harness; offered only inside GitHub Copilot.Compared with MAI-Thinking-1 Primary source[2]Training disclosure; Distribution channels [1]
  • Input text Primary source[2]Model Summary: Inputs
  • Output text Primary source[2]Model Summary: Outputs
  • Context window 256K tokens tokensThe numeric value reads K as 1,000; the source does not say. Primary source[2]Model Summary: Context length
  • Parameters 137B total, 5B activeSparse mixture-of-experts; the June launch post gives only the 5 billion active parameters. Primary source[2]Model Summary: Parameters [9]Our Models

Lineage

Predecessors

No known predecessor.

Successors

Based on

Variants and derived

None recorded.

Siblings

None recorded.

All ancestors

All descendants

Variants

No variants recorded in this record.

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from Microsoft on AI Radar →