---
title: "MAI Code 2 and Flash arrive in Copilot: access, benchmarks and the migration deadline"
description: "Compare MAI Code 2 and Flash benchmarks, GitHub Copilot plan eligibility, Foundry preview access and the December 18 Code 1.1 Flash migration deadline."
url: "https://promptszone.com/mai-code-2-flash-copilot-benchmarks-access"
published: "2026-11-17T23:00:05Z"
categories: ["New Models"]
---

# MAI Code 2 and Flash arrive in Copilot: access, benchmarks and the migration deadline

MAI Code 2 and Flash are available in GitHub Copilot. Compare three coding benchmarks, plan eligibility and the December 18 retirement of Code 1.1 Flash.

![Microsoft-reported pass rates: MAI Code 2 versus Flash scores 73.2% versus 59.9% on DeepSWE 1.1, 84.6% versus 78.3% on TerminalBench 2.1, and 43.1% versus 44.8% on SWE Atlas. Higher is better; each row's higher value is bold.](https://promptszone.com/assets/uploads/c6e20e672b528645fffb45a824e4b3ce.png)

**Summary:**

The larger model leads two of Microsoft's three published coding tests, while Flash leads SWE Atlas. Access and migration details matter as much as the headline cost claims.

Microsoft's MAI Code 2 and MAI Code 2 Flash reached GitHub Copilot on November 17. The launch offers two choices: a larger model for substantial repository changes and a lighter model for frequent edits. It also starts a migration clock for teams still using MAI-Code-1.1-Flash.

## Where you can use them

[GitHub's release notice](https://github.blog/changelog/2026-11-17-mai-code-2-and-mai-code-2-flash-in-github-copilot/) lists Code 2 for Copilot Pro, Pro+, Max, Business and Enterprise. Flash also includes Free and Student plans. Eligible annual-plan subscribers are included. Both use usage-based billing.

The model picker is available across supported surfaces including VS Code, Visual Studio, Copilot CLI, the cloud agent, GitHub's app and website, mobile, JetBrains, Xcode and Eclipse. Business and Enterprise access remains subject to administrator model policies.

Outside Copilot, Microsoft lists Code 2 Flash as a limited-access Foundry preview. OpenRouter, Vercel, Hermes, OpenCode and Kilo are described as upcoming destinations, not current launch access. That distinction is important if your workflow depends on a particular API provider.

## Three tests show why “larger” is not a universal answer

The November 17 [model card](https://microsoft.ai/pdf/MAI-Code-2-And-MAI-Code-2-Flash-Model-Card.pdf?v=2) reports the following pass rates. Higher is better; bold identifies the higher result in each row.

| Benchmark | MAI Code 2 | MAI Code 2 Flash |
| --- | --- | --- |
| DeepSWE 1.1 | **73.2%** | 59.9% |
| TerminalBench 2.1 | **84.6%** | 78.3% |
| SWE Atlas | 43.1% | **44.8%** |

Microsoft ran these tests on MSBench using GitHub Copilot CLI 1.0.102, maximum reasoning effort and no subagents. The figures average three runs per task and represent single-attempt pass rates, not the chance of succeeding after several attempts. Infrastructure errors count as failures.

Code 2 leads two rows, but Flash leads SWE Atlas. For a team choosing a default, that is a reason to test representative tasks rather than infer every outcome from model size. The card also warns that long sessions can repeat calls, drift from instructions or end without fully verifying work.

## Read the cost comparison with its promotional terms

In its [launch analysis](https://microsoft.ai/news/introducing-mai-code-2-and-mai-code-2-flash/), Microsoft reports that Code 2 reaches 98% of GPT-6.1 Sol's performance at 40% of the cost across four coding benchmarks. The MAI cost calculation uses a 50% promotional price. That is a workload-specific comparison, not a promise that every Copilot bill will fall by the same amount.

Our suggested pilot is a small set of recent bug fixes and multi-file changes with known acceptance tests. Compare successful merges, elapsed time, usage charges and reviewer correction time. Keep the original tests intact, and run the same tasks with each candidate. Our [guide to independent agent verification](https://promptszone.com/google-antigravity-teamwork-research-verification) explains why separating implementation from acceptance checks is useful.

## Existing Flash users have a December deadline

[GitHub will retire MAI-Code-1.1-Flash on December 18, 2026](https://github.blog/changelog/2026-11-17-upcoming-deprecation-of-mai-code-1-1-flash-in-december/), across Copilot experiences. It recommends moving to the latest MAI Code Flash model before then. Administrators should confirm the replacement is enabled before teams update their workflows.

## Sources and image credit

Sources: GitHub's November 17 availability and retirement notices; Microsoft's November 17 launch article and model card, linked above. Chart: PromptsZone, using the model card's developer-reported results. The official Microsoft provider logo comes from the [Microsoft News Center](https://news.microsoft.com/microsoft-logo_rgb_c-gray-2/) under its [editorial-use permission](https://news.microsoft.com/en-sg/default/); it is a provider-logo fallback, not a distinct MAI Code 2 logo.
