All news

AI News7 min readBy JH Akash

MAI Code 2: 98% of GPT-6.1 Sol at 40% of the Price

Microsoft launched seven MAI models at Ignite 2026. MAI Code 2 claims 98% of GPT-6.1 Sol's coding quality at 40% of the price, and 96% of Claude Opus 5.5 at 5% of the cost. Here is what launched, what it costs, and the catch.

MAI Code 2 vs GPT-6.1 Sol infographic: 98% performance at 40% of the price, Microsoft 7 new MAI models at Ignite 2026

Microsoft just fired its biggest shot yet at OpenAI and Anthropic. At Ignite 2026 on November 17, Microsoft AI launched seven new MAI models in one go, and the pitch is not raw power. It is price. The headline model, MAI Code 2, claims 98% of GPT-6.1 Sol's coding quality at about 40% of the price, and 96% of Claude Opus 5.5 at just 5% of the cost. Both coding models are already generally available in GitHub Copilot.

The catch, and it matters, is that every number below is vendor-reported. Microsoft picked the benchmarks, ran most of the tests itself, and priced its models at a 50% promotional discount. This post lays out what launched, what it costs, and where the claims get thin.

Quick verdict

QuestionShort answer
What launched?Seven MAI models: Code 2, Code 2 Flash, Cyber 2 Flash, Image 2.7, Realtime 1, Transcribe 2, Decision 1.1
The headline claimMAI Code 2: 98% of GPT-6.1 Sol, 40% of the price
The boldest claimMAI Cyber 2 Flash: GPT-6 Astra quality at under 2% of the cost per task
Where to try itMAI Code 2 and Code 2 Flash are live in GitHub Copilot now
The catch50% promo pricing, doubles after May 17, 2027; vendor-chosen benchmarks

The seven MAI models at a glance

ModelWhat it doesHeadline claim (Microsoft-reported)Availability
MAI Code 2Everyday enterprise coding98% of GPT-6.1 Sol at 40% of the price; 96% of Claude Opus 5.5 at 5% of the costGenerally available in GitHub Copilot
MAI Code 2 FlashFast, high-volume codingMatches GPT-6 Luna at 65% of the price; matches Claude Haiku 5.5 at 20%GitHub Copilot, including Free and Student plans
MAI Cyber 2 FlashFinding software vulnerabilitiesGPT-6 Astra quality at under 2% of the cost per task on CyberGymE2EVerified defenders only, through MDASH
MAI Image 2.7Image generation and editingNo. 1 on Artificial Analysis image editing, at one fifth the price of GPT Image 2.6 Sunburst (max)MAI Labs, limited preview
MAI Realtime 1Speech-to-speech voiceOutperforms GPT-Realtime 2.1, from $0.01 per audio minutePreview
MAI Transcribe 2Multilingual transcription, behind Dragon Copilot TranscribePowers 10 million patient encounters a month, multilingual error rates halvedIn Dragon Copilot
Decision 1.1Model routing inside apps50x faster and 79x cheaper than GPT-6.1 SolGenerally available

Microsoft also announced Maia 200, its second-generation AI accelerator, now in production and delivering 1.7x more tokens per watt. And with MAI Licensed Weights, companies can take a frontier MAI model, customize it with their own data, and run it inside their own Azure environment.

MAI Code 2 vs GPT-6.1 Sol vs Claude Opus 5.5: benchmarks

Microsoft averaged four benchmarks aimed at everyday enterprise coding: DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, and FrontierCode v1.1. Three of the four ran on Microsoft's own evaluation platform; the fourth used results from Artificial Analysis. Here is the average score Microsoft reported.

Bar chart of coding performance, average of four benchmarks: MAI Code 2 63.2 percent, GPT-6.1 Sol 64.1 percent, Claude Opus 5.5 65.3 percent

ModelAverage score across 4 benchmarksShare of GPT-6.1 Sol performanceShare of Claude Opus 5.5 performance
MAI Code 263.2%98%96%
GPT-6.1 Sol64.1%100%98%
Claude Opus 5.565.3%102%100%

Read the chart honestly: the three models sit within about two points of each other. MAI Code 2 is not the best coder here. It is the cheapest by a wide margin, and that is the whole pitch. An average can also hide a benchmark where Code 2 trails badly, and Microsoft has not published the per-benchmark breakdown in the announcement.

Pricing: where the savings really come from

Bar chart of cost per coding task at promo price: MAI Code 2 0.31 dollars, GPT-6.1 Sol 0.76 dollars, Claude Opus 5.5 6.31 dollars

ModelInput price per 1M tokensOutput price per 1M tokensCost per coding task (promo)
MAI Code 2$0.50$2.50$0.31
MAI Code 2 Flash$0.05$0.25Not published
GPT-6.1 Sol$2.00$10.00$0.76
Claude Opus 5.5$4.00$20.00$6.31

The per-task costs above are at MAI Code 2's 50% promotional price. At full price, Code 2 would cost roughly four-fifths as much as GPT-6.1 Sol per task and about a tenth as much as Claude Opus 5.5. The promotional prices double after May 17, 2027, so the 40% and 5% figures are time-limited. Still, even at list price the gap is large, and that is what makes this launch matter for teams running coding agents all day.

For image work, Microsoft quotes about $35 per 1,000 images for MAI Image 2.7 and $10 for the Flash version. MAI Realtime 1 starts at $0.01 per audio minute.

The security play: MAI Cyber 2 Flash

The most striking number of the launch is not about coding. On CyberGymE2E, a benchmark of real-world vulnerability discovery, Microsoft says MAI Cyber 2 Flash completed 92.3% of tasks against 91.5% for GPT-6 Astra, at less than 2% of the cost per task. In other words, slightly better quality for roughly one-fiftieth of the price, on Microsoft's own measurement.

There is a limit, though. Cyber 2 Flash is not open to everyone. Microsoft is making it available initially to a limited group of verified defenders, so most developers cannot try it yet. OpenAI, meanwhile, launched its own security push the same week with Daybreak Red Team, fleets of agents that pen-test web applications.

What Microsoft is really doing

This launch is about independence. Microsoft has spent years as OpenAI's biggest partner and customer. Now it is building its own frontier models, its own Maia 200 chips to run them, and a licensing model that lets enterprises customize and self-host. Microsoft AI chief Mustafa Suleyman framed it as a response to what CIOs and CFOs keep telling him: token spending is soaring, and agents that run 24 hours a day need cheaper models.

Note the timing too. OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol on the same day, November 17, which means Microsoft's comparison against GPT-6.1 Sol is against the previous generation. The new GPT-6.2 Astra costs $8 per million input tokens and $40 per million output, down 20% from GPT-6 Astra, so OpenAI is cutting prices at the same time.

Which to pick

  • Pick MAI Code 2 if your team lives in GitHub Copilot and runs lots of everyday coding tasks: debugging, multi-file changes, refactoring, and shipping code to production. The savings compound fast at high volume.
  • Pick MAI Code 2 Flash for quick explanations, targeted edits, and routine fixes, especially on Copilot Free or Student plans where it is already included.
  • Stick with GPT-6.1 Sol or Claude Opus 5.5 if you need the strongest general reasoning, frontier research, or very long-horizon agent work. Microsoft says Code 2 is explicitly not yet tuned for those.
  • Watch MAI Cyber 2 Flash if you run a security team, but expect a wait unless you are a verified defender.
  • Try MAI Image 2.7 in MAI Labs if image editing cost is your bottleneck. The No. 1 Artificial Analysis ranking is an independent leaderboard, which makes it the most independently verified claim in the launch.

If you are building AI agents for your business and want help choosing the right model stack, read our guide on AI agent development services or talk to us about hiring an AI automation agency. For the wider model race, see our GPT-6 Astra breakdown and our Claude Opus 5.5 overview.

FAQ

What is MAI Code 2?

MAI Code 2 is Microsoft's flagship coding model, launched November 17, 2026 at Ignite. It is built for everyday enterprise coding work such as debugging, multi-file changes, and refactoring, and it is generally available in GitHub Copilot.

How much does MAI Code 2 cost?

At its 50% promotional price, MAI Code 2 costs $0.50 per million input tokens and $2.50 per million output tokens. Microsoft says the promotional prices double after May 17, 2027.

Is MAI Code 2 better than GPT-6.1 Sol?

Not on raw scores. Microsoft reports MAI Code 2 at 63.2% averaged across four coding benchmarks, against 64.1% for GPT-6.1 Sol and 65.3% for Claude Opus 5.5. Code 2 wins on price, not on quality, and the benchmarks were chosen and mostly run by Microsoft.

Is MAI Code 2 available in GitHub Copilot?

Yes. MAI Code 2 and MAI Code 2 Flash are generally available in GitHub Copilot, in the model picker and through Copilot Fusion. Code 2 Flash is also available on Copilot Free and Student plans.

What are the seven MAI models?

MAI Code 2, MAI Code 2 Flash, MAI Cyber 2 Flash, MAI Image 2.7, MAI Realtime 1, MAI Transcribe 2 (behind Dragon Copilot Transcribe), and Decision 1.1. Microsoft also put its Maia 200 accelerator into production and launched MAI Licensed Weights for enterprise customization.

What is the catch with Microsoft's price claims?

Three things. First, the cost figures use a 50% promotional price that doubles after May 17, 2027. Second, Microsoft picked the benchmarks and ran three of the four itself, with no independent verification yet. Third, Code 2 is not tuned for frontier research or very long-horizon agent work, and the GPT-6.1 Sol comparison is against the previous generation, since GPT-6.2 launched the same day.

  • MAI Code 2
  • Microsoft
  • GitHub Copilot
  • GPT-6.1 Sol
  • Claude Opus 5.5
ChatGPT Wallet infographic: AI agents buy, Face ID approves, launched Nov 13, US Pro only, no card number shared

AI News

ChatGPT Wallet: AI Agents Buy, Face ID Approves

OpenAI's ChatGPT Wallet, launched Nov 13, lets dots agents check out for you without ever seeing your card number. Every purchase still needs your Face ID. Here is how it works, who can use it, and what it costs.

n8n Nod video thumbnail: free open-source personal AI agent

AI

n8n Nod: Free Open-Source AI Agent on Your PC

n8n's Nod is the free, open-source personal AI agent that runs on your own computer with any model you pick. How it works, how to install it in 7 steps, and how it stacks up against Dots, Duet, and Grokbot.