Subscribe
0%
Illustration for: Microsoft's MAI Code 2 lands in Copilot, priced against Claude and GPT: what its Ignite claims show

Microsoft's MAI Code 2 lands in Copilot, priced against Claude and GPT: what its Ignite claims show

At Ignite on November 17, Microsoft AI launched MAI Code 2, Code 2 Flash, a new security model and several others, all pitched on cost. Microsoft says Code 2 gets close to GPT-6.1 Sol and Claude Opus 5.5 for a fraction of the price. Every number is Microsoft's own, and the coding prices assume a 50% promotion with no stated end date.

Listen to this article0:00 / 4:54

Microsoft AI used the first day of Ignite on November 17 to launch a batch of in-house models, led by MAI Code 2, a coding model now selectable in GitHub Copilot. Microsoft's pitch is price: it says Code 2 delivers about 98% of OpenAI's GPT-6.1 Sol and 96% of Anthropic's Claude Opus 5.5 on its chosen coding benchmarks, at roughly 40% and 5% of their cost respectively.

Why it matters: Microsoft is now pricing its own models directly against OpenAI's and Anthropic's inside its own developer tools, a move Reuters framed as taking on both labs. Microsoft AI chief Mustafa Suleyman said enterprise customers say "token spend is going through the roof," according to Yahoo Finance. If the cost claims hold up in real codebases, Copilot users get a cheaper default for everyday work. But none of the figures have been independently tested yet.

What was announced

Microsoft's Ignite post headlines "seven" models but lists more items. The main ones:

Model What it's for Availability (per Microsoft)
MAI Code 2 Multi-file changes, debugging, repo-wide reasoning, UI work GitHub Copilot
MAI Code 2 Flash High-volume routine edits and explanations GitHub Copilot; limited-access preview in Microsoft Foundry
MAI Cyber 2 Flash Vulnerability discovery and triage Update for existing customers of Microsoft's MDASH security system
MAI Image 2.7 / 2.7 Flash Image generation and editing Microsoft Foundry
MAI Realtime 1 Live speech-to-speech Try on MAI Labs; no general availability stated
Dragon Copilot Transcribe (on MAI Transcribe 2) Clinical transcription Live in Dragon Copilot
Decision 1.1 Routes requests to bigger models only when needed Generally available

Microsoft also announced MAI Licensed Weights, a way for customers to customise MAI models and run them in their own Azure environment, with no availability date.

The coding numbers, and their conditions

Microsoft averages four benchmarks for Code 2 (DeepSWE 1.1, Terminal-Bench 2.1, SWE-Atlas and FrontierCode v1.1), all at maximum reasoning effort. Unite.AI and Securities.io report the underlying figures:

Model Avg. pass rate Avg. cost per task
MAI Code 2 63.2% $0.309
GPT-6.1 Sol 64.1% $0.761
Claude Opus 5.5 65.3% $6.306

For Code 2 Flash, averaged over three of those benchmarks, Unite.AI reports 61.0% at $0.052 per task, against 55.1% for GPT-6 Luna and 53.9% for Claude Haiku 5.5. Microsoft's post describes Flash as "comparable" to both, at about 65% of Luna's price and 20% of Haiku's.

Four caveats sit underneath those tables:

  • Promotional pricing. Per Unite.AI and DataStudios, Microsoft's cost figures assume a 50% promotional price, and the announcement doesn't say when it ends. At list price, the gap narrows.
  • Microsoft's own harness. Three of the benchmarks were run on Microsoft's internal MSBench platform through GitHub Copilot CLI, without subagents, according to Unite.AI.
  • Stated limits. Microsoft says the Code 2 models are not tuned for fundamental science research or very long-horizon agentic work.
  • Small inconsistencies. The blind frontend preference test is reported as 67% in one Microsoft description and 68% on the Ignite page.

The security model

Microsoft says MAI Cyber 2 Flash matches GPT-6 Astra on the CyberGymE2E benchmark at less than 2% of the cost per task, and that its output tokens cost 86% less than the previous MAI Cyber 1.1 Flash. Securities.io reports 92.3% task success at $0.45 per task versus 91.5% at $34.38 for GPT-6 Astra, on 813 of 905 tasks after Microsoft excluded 92. It also reports a one-million-token context window and automatic delivery to eligible customers from mid-December. Another write-up cites an 83.2% score instead, so treat the exact figure as unconfirmed until Microsoft publishes its methodology. Reuters reports the model will go to a limited group of customers.

Who gets it in Copilot

DataStudios, citing GitHub's changelog, reports that Code 2 Flash is available on every Copilot plan including Free and Student, while full Code 2 requires Pro, Pro+, Max, Business or Enterprise. Organisation access also depends on admins' model policies. As of launch day, GitHub's per-token rates for the new models hadn't appeared on its official pricing list, according to the same report.

What to do

  • Copilot users: try Code 2 Flash on routine edits and Code 2 on a real multi-file task you've already solved with another model. Compare results and credit use. Your own repo is a better test than Microsoft's averages.
  • Copilot admins: check whether your model policy allows the MAI models before developers ask, and watch for the end of the promotional pricing.
  • Everyone else: wait for independent results. Every benchmark and price comparison here is Microsoft's own, run on benchmarks Microsoft chose.

Sources

  1. Microsoft AI — Seven new world class models at scalable price-performance microsoft.ai
  2. Reuters via The Korea Times — Microsoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAI koreatimes.co.kr
  3. Yahoo Finance — Microsoft's AI pitch at Ignite: Cheaper models as business 'token spend is going through the roof' finance.yahoo.com
  4. Unite.AI — Microsoft AI Releases MAI Code 2 and MAI Code 2 Flash for Enterprises unite.ai
  5. Securities.io — Microsoft Releases Seven AI Models With Lower-Cost Enterprise Options securities.io
  6. DataStudios — Microsoft MAI Code 2 and Code 2 Flash in GitHub Copilot: plans, benchmarks, pricing datastudios.org

#Microsoft#MAI Code 2#MAI Cyber 2 Flash#GitHub Copilot#Microsoft Foundry#Ignite 2026#coding models#Mustafa Suleyman