Microsoft's 5%-of-Claude-Opus-5.5 cost claim for MAI Code 2 rests on a GitHub half price that ends on 17 May 2027
Microsoft launched MAI Code 2 and MAI Code 2 Flash at Ignite on 17 November 2026, claiming 96% of Claude Opus 5.5's coding performance "at just 5% of the cost". A footnote in Microsoft's own post says the cost is "shown at its 50% promotional price". GitHub's Copilot pricing page, as of 18 November 2026, gives the end date: MAI Code 2 costs $0.50 per million input tokens and $2.50 per million output tokens "until May 17, 2027", then $1 / $5. On that day, by my arithmetic, every cost ratio Microsoft published doubles if the models spend the same tokens per task and competitors keep their 18 November 2026 prices, and MAI Code 2 Flash goes from 65% of GPT-6 Luna's cost to about 130%.
What Microsoft measured
Microsoft averages four benchmarks "that target the everyday enterprise coding tasks we see in GitHub Copilot": DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1. Three ran on MSBench, Microsoft's internal evaluation platform, through the GitHub Copilot CLI at max reasoning effort; FrontierCode figures come from Artificial Analysis. On that average, Microsoft reports 98% of GPT-6.1 Sol's performance at 40% of the cost, and 96% of Claude Opus 5.5's at 5%. For MAI Code 2 Flash it claims "comparable performance to GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of the cost" on three of the benchmarks, also at the promotional price.
GitHub Copilot prices, as of 18 November 2026
US dollars per million tokens, input / output, from GitHub's models and pricing page. Footnotes on the page give a dated full price for both new MAI models; Gemini 4 Carbon, discussed further down, has one too.
| Model | Price now | From 17 May 2027 |
|---|---|---|
| Microsoft MAI Code 2 | $0.50 / $2.50 | $1 / $5 |
| Microsoft MAI Code 2 Flash | $0.05 / $0.25 | $0.10 / $0.50 |
| Microsoft MAI-Code-1.1-Flash | $0.20 / $1.20 | $0.20 / $1.20 |
| Anthropic Claude Opus 5.5 | $4 / $20 | $4 / $20 |
| OpenAI GPT-6.1 Sol | $2 / $10 | $2 / $10 |
| Anthropic Claude Haiku 5.5 (prompts up to 100,000 tokens) | $0.10 / $0.50 | $0.10 / $0.50 |
| OpenAI GPT-6 Luna (prompts up to 272,000 tokens) | $0.10 / $0.50 | $0.10 / $0.50 |
Per token, MAI Code 2 costs 12.5% of Claude Opus 5.5 today and 25% from 17 May 2027. Anthropic's own pricing page, as of 18 November 2026, lists Opus 5.5 at the same $4 / $20. Microsoft's 5% sits below the per-token 12.5%, which in my reading means MAI Code 2 spent fewer tokens per task than Opus 5.5 on these benchmarks.
MAI Code 2 Flash is the sharper case. Its full price, $0.10 / $0.50, is exactly what GitHub charges for GPT-6 Luna and Claude Haiku 5.5. It still costs less than MAI-Code-1.1-Flash, Microsoft's previous small coding model, announced on 11 August 2026, which GitHub lists at $0.20 / $1.20.
Microsoft's cost ratios at the full price
Microsoft's footnotes say its own models' costs are shown at the promotional price; nothing says the competitors' are discounted. If the models use the same tokens per task, doubling the MAI price doubles each ratio. The right column is my arithmetic.
| Claim | Microsoft, at promotional price | At the 17 May 2027 price |
|---|---|---|
| MAI Code 2 vs Claude Opus 5.5 | 5% of the cost | about 10% |
| MAI Code 2 vs GPT-6.1 Sol | 40% of the cost | about 80% |
| MAI Code 2 Flash vs Claude Haiku 5.5 | 20% of the cost | about 40% |
| MAI Code 2 Flash vs GPT-6 Luna | 65% of the cost | about 130% |
Why a launch price: my reading, and the other one
Microsoft owns both the model and the shop. The benchmarks were run through GitHub's own Copilot CLI, and GitHub's changelog says MAI Code 2 Flash is offered on the Copilot Free plan and that new models are enabled automatically unless an administrator turns them off. My reading is that six months at half price is meant to make MAI the default habit in Copilot before the full price applies.
The other explanation is that launch prices are routine and Microsoft disclosed the basis itself. The same GitHub page shows Google's Gemini 4 Carbon at $2 / $10 until 1 December 2026 and $4 / $20 after, a doubling I covered on 17 November 2026. Even at $1 / $5, MAI Code 2 costs a quarter of Opus 5.5 per token.
For a team choosing a default Copilot model, the number to test is cost per task at $1 / $5 and $0.10 / $0.50, not the launch figure.
My forecast: on 18 May 2027, GitHub's models and pricing page will list MAI Code 2 Flash, or a newer MAI Code Flash model shown in its place, below $0.10 / $0.50 on input, output or both, through an extended promotion or a lower list price. My reason: at full price Flash costs more than GPT-6 Luna per task, by my arithmetic from Microsoft's 65% figure, so it loses its price argument against OpenAI's small model. If the cheapest MAI Code Flash model on that page costs $0.10 / $0.50 or more that day, or the page still shows the promotional price as ending on 17 May 2027, I was wrong.