MAI Code 2 vs Claude Opus 5.5 and GPT-6.1 Sol: $0.50 Input vs $4, Promo Price Until May 2027
Microsoft released MAI Code 2 at Ignite 2026 on Nov 17, and it entered GitHub Copilot the same day; Microsoft says it reaches 96% of Claude Opus 5.5's and 98% of GPT-6.1 Sol's performance across four coding benchmarks, at 5% and 40% of their per-task costs. This comparison checks the three coding flagships item by item — positioning, specs, pricing and benchmarks — against official spec pages, pricing pages and public leaderboards, with each figure sourced and dated; Microsoft's comparison figures come from its own test platform, and the costs are calculated at promotional prices.

Latest
1 updates · newest first-
Comparison published: MAI Code 2 vs Claude Opus 5.5 and GPT-6.1 Sol is live.
Further updates will appear here
Side by side
✓ marks the side that wins the row| Item | MAI Code 2 | Claude Opus 5.5 / GPT-6.1 Sol |
|---|---|---|
| Price | Input $0.50, output $2.50, cached input $0.025 (per 1M tokens, promotional until May 17, 2027)✓ | Opus 5.5: input $4.00, output $20.00; GPT-6.1 Sol: input $2.00, output $10.00 (per 1M tokens) |
| Context | 1,050,000 tokens | Opus 5.5: 1,000,000 tokens; GPT-6.1 Sol: 1,050,000 tokens |
| Microsoft four-benchmark average | 63.2% | Opus 5.5: 65.3%; GPT-6.1 Sol: 64.1%✓ |
| Cost per task (Microsoft) | $0.309✓ | Opus 5.5: $6.306; GPT-6.1 Sol: $0.761 |
| Intelligence Index | Not published | Opus 5.5: 58; GPT-6.1 Sol: 55 (Artificial Analysis v5)✓ |
| Terminal coding | Not published | Opus 5.5: 62 (Coding Agent Index); GPT-6.1 Sol: not published |
| Standalone API | No, available only inside GitHub Copilot | Opus 5.5 and GPT-6.1 Sol both offer public APIs✓ |
Verdict
Verdict: input costs $0.50 against $4 and $2, output $2.50 against $20 and $10; Microsoft's four-benchmark average pass rates are 63.2%, 65.3% and 64.1%; Opus 5.5 leads on the independent leaderboard, where MAI Code 2 has no row yet; the promotional price ends May 17, 2027, after which input and output rates double.
01Overview
The three models shipped within two months of each other, each as its maker's coding flagship: MAI Code 2 is the coding model Microsoft launched at Ignite 2026, Claude Opus 5.5 is the flagship of Anthropic's 5.5 family from Sep 22, and GPT-6.1 Sol is OpenAI's coding and professional-work model from Sep 29. The price table spans two orders of magnitude: MAI Code 2 at $0.50 input, GPT-6.1 Sol at $2, Opus 5.5 at $4; the dividing lines are access and price validity — MAI Code 2 is available only inside GitHub Copilot and at a promotional price, while the other two have public APIs.
The timing matters: OpenAI launched GPT-6.2 Sol on the same day MAI Code 2 shipped, keeping GPT-6.1 Sol's $2 and $10 pricing, while Microsoft's comparison still targets the previous generation, GPT-6.1 Sol; on the Artificial Analysis Intelligence Index v5, Opus 5.5 scores 58 at max effort and GPT-6.1 Sol 55, while MAI Code 2 has no independent leaderboard row yet.
- Publisher and date: Microsoft; MAI Code 2 released at Ignite 2026 on Nov 17, 2026, and entered GitHub Copilot the same day
- Publishers and dates: Anthropic released Claude Opus 5.5 on Sep 22, 2026; OpenAI released GPT-6.1 Sol on Sep 29, 2026
- Method: specs and prices from each vendor's official pages and GitHub's Copilot billing documentation (checked Nov 19, 2026); benchmarks split into Microsoft-reported figures and Artificial Analysis independent measurements; three of Microsoft's four benchmarks ran on its own MSBench platform, with the FrontierCode figure cited from Artificial Analysis
- Context: OpenAI released GPT-6.2 Sol on the same day, Nov 17, keeping GPT-6.1 Sol's $2 and $10 pricing; Microsoft's comparison still targets the previous generation, GPT-6.1 Sol
02Specs and pricing
On specs, all three models sit at the 1-million-token context scale with the same 128,000-token output cap; only one publishes a parameter count — MAI Code 2 at roughly 2.8 trillion total parameters with 77 billion active per token, while the other two publish none. On price, input runs $0.50 against $2 against $4, and output $2.50 against $10 against $20; the dividing lines are price validity and access — MAI Code 2's current price is promotional, doubling from May 17, 2027, and it shipped with no standalone API.
- Scale: MAI Code 2 has roughly 2.8 trillion total parameters with 77 billion active per token, plus 3 billion embedding parameters (model card); Claude Opus 5.5 and GPT-6.1 Sol publish no parameter counts
- Context and output caps: MAI Code 2 has a 1,050,000-token context window, 922,000-token maximum input and 128,000-token output cap; Opus 5.5 has a 1,000,000-token context window and 128,000-token output cap; GPT-6.1 Sol has a 1,050,000-token context window and 128,000-token output cap
- Input modalities: MAI Code 2 takes text and image input and returns text; Opus 5.5 takes text, image and file input; GPT-6.1 Sol takes text and image input and returns text
- API pricing (per 1M tokens): MAI Code 2 at $0.50 input, $0.025 cached input and $2.50 output, promotional until May 17, 2027, then $1.00 input, $0.05 cached input and $5.00 output (GitHub Copilot billing documentation); Opus 5.5 at $4.00 input, $0.20 cache read and $20.00 output, with the Batch API at $2.00 input and $10.00 output and fast mode at $8.00 input and $40.00 output (Anthropic pricing documentation); GPT-6.1 Sol at $2.00 input, $0.10 cached input and $10.00 output, with requests over 272,000 input tokens billed at $4.00 input and $15.00 output for the whole request (OpenAI pricing page)
- Price ratios: at published list prices, MAI Code 2's promotional input and output rates are 12.5% of Opus 5.5's and 25% of GPT-6.1 Sol's; at the standard rates from May 17, 2027, they are 25% of Opus 5.5's and 50% of GPT-6.1 Sol's (calculated from published list prices)
- Plans and access: MAI Code 2 is available inside GitHub Copilot on Copilot Pro, Pro+, Max, Business and Enterprise, with no standalone API at launch, while Flash has a limited Foundry preview; Opus 5.5 is available via the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry, with Claude Pro at $20 a month, Max at $100 and Ultra at $200, and the Free plan excludes Opus; GPT-6.1 Sol is available via the OpenAI API, ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu
- Openness: all three are closed and hosted, with no public weights; MAI Code 2 offers licensed weights (MAI Licensed Weights, customized by enterprises and run in their own Azure environment), which are not open weights
- Knowledge cutoffs: MAI Code 2, August 2026; Opus 5.5, June 2026; GPT-6.1 Sol, April 30, 2026
03Benchmarks
Read the scores in two groups: Microsoft's four-benchmark comparison is the vendor's own runs on its own harness, so only the same benchmark at the same version sits side by side; the Artificial Analysis Intelligence Index v5 is one institution measuring Opus 5.5 and GPT-6.1 Sol on the same protocol, while MAI Code 2 has no independent row on that leaderboard yet. The two groups do not combine into one ranking table.
- Microsoft's four-benchmark coding average pass rate (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, FrontierCode v1.1, published by Microsoft on Nov 17, 2026): MAI Code 2 63.2%, GPT-6.1 Sol 64.1%, Claude Opus 5.5 65.3%, within 2.1 points of each other; three of the four ran on Microsoft's own MSBench platform, with the FrontierCode figure cited from Artificial Analysis; all three models ran at maximum reasoning effort
- Average cost per task across Microsoft's four benchmarks (calculated at promotional prices, published by Microsoft): MAI Code 2 $0.309, GPT-6.1 Sol $0.761, Claude Opus 5.5 $6.306; Opus 5.5 costs about 20 times as much per task as MAI Code 2
- Artificial Analysis Intelligence Index v5 (released Oct 27, 2026, max effort): Claude Opus 5.5 58, GPT-6.1 Sol 55, a 3-point gap; MAI Code 2 has no independent row on the leaderboard and publishes no score
- Cost per Intelligence Index task (v5, max effort, Artificial Analysis): Claude Opus 5.5 $10.04, GPT-6.1 Sol $1.27; MAI Code 2 publishes no figure
- Artificial Analysis Coding Agent Index: Claude Opus 5.5 62 (via Claude Code); GPT-6.1 Sol and MAI Code 2 publish no score for this index
- FrontierCode v1.1 (Artificial Analysis): Claude Opus 5.5 56%; GPT-6.1 Sol and MAI Code 2 publish no standalone score for this benchmark, and MAI Code 2's figure appears only inside Microsoft's four-benchmark average
- Front-end blind evaluation (published by Microsoft, 258 annotators): in blind side-by-side evaluations of front-end work — interfaces, games, data visualizations, 3D scenes and SVG graphics — developers preferred MAI Code 2's output 64% of the time, against GPT-6 Sol (Microsoft's wording); this is a preference study, not a pass rate
- Successor check: GPT-6.2 Sol, the successor to GPT-6.1 Sol, launched Nov 17 with an Artificial Analysis Intelligence Index score of 57.3 and the same $2 and $10 pricing; Microsoft's comparison still uses GPT-6.1 Sol
04How to choose
Match the scenario to the numbers on function and price: pay-as-you-go unit prices, long context and standalone API access each point to a different model in the table; self-hosted weights and video input are outside what any of the three models offers.
- Lowest pay-as-you-go token price: MAI Code 2 at $0.50 input and $2.50 output per million tokens on promotion until May 17, 2027, then $1.00 input and $5.00 output; GPT-6.1 Sol at $2.00 input and $10.00 output; Opus 5.5 at $4.00 input and $20.00 output
- Standalone API access: Opus 5.5 and GPT-6.1 Sol offer public APIs; MAI Code 2 is available only inside GitHub Copilot at launch with no standalone API, while Flash has a limited Foundry preview
- Longest context: MAI Code 2 and GPT-6.1 Sol at 1,050,000 tokens, Opus 5.5 at 1,000,000 tokens; all three cap single responses at 128,000 tokens
- Measured terminal-coding scores: on the Artificial Analysis Coding Agent Index, Opus 5.5 scores 62 (via Claude Code); GPT-6.1 Sol and MAI Code 2 publish no score for this index; on Microsoft's own four-benchmark average the three score 63.2%, 64.1% and 65.3%
- Low cached-input price: MAI Code 2 at $0.025, GPT-6.1 Sol at $0.10 and Opus 5.5 at $0.20 per million tokens for cache reads
- Use inside Copilot: MAI Code 2 is available on Copilot Pro, Pro+, Max, Business and Enterprise; Opus 5.5 and GPT-6.1 Sol need Copilot Pro+ or above inside Copilot
- Self-hosted weights: none of the three models publishes weights; MAI Code 2 offers licensed weights (customized by enterprises and run in their own Azure environment), which are not open weights
- Science research or very long agentic tasks: Microsoft says MAI Code 2 is not yet tuned for fundamental science research or very long-horizon agentic work, and that kind of task sits outside the four benchmarks in this comparison
Why it matters
MAI Code 2 charges $0.50 per million input tokens, with a 63.2% four-benchmark average within 2.1 points of its rivals; the promotional price ends May 17, 2027, after which Copilot usage rates double.
CompaniesOpenAI·Anthropic·Microsoft
SourcesMicrosoft's seven MAI models announcement (Nov 17, 2026); Microsoft's MAI Code 2 and Code 2 Flash introduction (Nov 17, 2026); GitHub Copilot models and pricing documentation (checked Nov 18, 2026); Anthropic pricing page (checked Nov 19, 2026); OpenAI API pricing page (checked Nov 19, 2026); Artificial Analysis Intelligence Index v5 coverage (Oct 27, 2026); OrcaRouter's MAI Code 2 model-card rundown (Nov 18, 2026); HIMIAO's Microsoft MAI story (Nov 18, 2026); HIMIAO's OpenAI GPT-6.2 flash (Nov 17, 2026). This article compiles public information and does not constitute investment advice.
From HIMIAO Horizon · himiao.ai
Follow HIMIAO Horizon
No sign-up · Unsubscribe any timeNotifications
A ping on your phone or computer when the brief is out. No email needed.
Comments
Related
- Nov 17Microsoft Unveils Seven MAI Models: Code 2 Matches Claude at 5% of the Price
- Nov 15SemiAnalysis Names Nvidia's Seven Biggest GPU Customers; Six Build Their Own Chips
- Nov 9Four AI Giants Back a Common Opt-Out Signal for AI Training
- Oct 30Four AI Labs Back a Shared Training Opt-Out Signal
- Oct 25Sweden's Government Offices Block ChatGPT and Claude on Work Computers From Oct 26
Today
Nov 19 Thursday- Brief
- GitHubGitHub Weekly Vol.6: 7 Projects Worth a Look This Week
Tested or read — weeklyAll →
- Quiz
- Call
Latest news
All →- Nov 18Study: Frontier Models Can Be Tricked Into Marking Civilian Targets
- Nov 18Nvidia Reaffirms Open-Source Strategy, Hugging Face Deal on Track
- Nov 17Klarna Launches AI Assistant That Shops, Books and Runs Errands
- Nov 18Palo Alto to Buy Applied Compute for $4.5 Billion in All-Stock Deal
- Nov 18NGINX2Root Exposed: Five 9.2-Severity Flaws Let Attackers Run Code on Nginx Without Logging In
- Nov 18Nvidia Stops Shipping Licensed Hopper Chips to China
™