| published by | Frederic Lardinois |
|---|---|
| in blog | The New Stack |
| published date | 2026-11-17 |
| original entry | Microsoft’s new coding model promises near-Opus performance at 5% of the cost |
At its Ignite conference in San Francisco on Tuesday, Microsoft announced a number of new AI models, including two new coding models.
Imaginatively named Code 2 and Code 2 Flash, these new models promise to come very close to models like GPT-6.1 Sol and Claude Opus 5.5 at less than half the cost.
As Microsoft notes, “MAI Code 2 is at the frontier of price-performance” for enterprise coding tasks. “It does all of this while also delivering changes code maintainers would merge, at a fraction of frontier cost.”
The promise here is that Code 2 gets to 98% of the performance of GPT-6.1 Sol at 40% of the cost and 96% of Claude Opus 5.5 at 5% of the cost (which also goes to show how expensive Anthropic’s models are to run right now).
The model, Microsoft says, has also been tuned to create “great dashboards and visualizations” and that it “excels at visual coding.”
At least for the OpenAI comparison, it’s worth noting that OpenAI today released GPT-6.2 Astra and Sol, with improved performance. That softens Microsoft’s claims a bit, but it’s worth noting that OpenAI, too, notes that at least on Agents’ Last Exam, GPT-6.2 Astra is now 5.5 points “ahead of Claude Fable 5.2 at just 5% of the cost” (which also goes to show how expensive Anthropic’s models are right now).

As for MAI Code 2 Flash, Microsoft argues that it can match Anthropic’s Claude Haiku 5.5 and GPT-6 Luna on benchmarks like DeepSWE 1.1, Terminal-Bench 2.1, and SWE Atlas at 20% and 65% of the cost, respectively.
One interesting wrinkle here is that Microsoft specifically tuned these models for the day-to-day coding work of its users. “They’re explicitly not yet tuned for harder problems, like fundamental science research and very long horizon agentic work,” Microsoft writes.
The models Microsoft is comparing these models to are generalists. For a while, OpenAI took a similar approach with its Codex models, but it has abandoned this approach in favor of more general models.
Microsoft, meanwhile, which doesn’t build those kinds of in-house frontier models, seems to be focused on smaller specialized models that it can run more efficiently. And while they won’t match the frontier models across every benchmark, focusing on these specific tasks makes a lot of sense in this context.
Microsoft is taking this even further with models like MAI Work Flash, a model trained to work across Microsoft 365 and Fabric IQ products.

MAI Code 2 and Code 2 Flash are now available in GitHub Copilot, while Code 2 Flash is in limited preview on Microsoft Foundry. OpenRouter, Vercel, Hermes, Open Code, and Kilo will soon get support as well.
AI Engineer Step by step guide to becoming an AI Engineer in 2026 Introduction Working With LLMs Context Engineering AI Models Embeddings & Vector Databases Embedding Models RAGs AI Agents AI Safety and Ethics Evaluation & ObservabiityYOUTUBE.COM/THENEWSTACK
Tech moves fast, don't miss an episode. Subscribe to our YouTube channel to stream all our podcasts, interviews, demos, and more.