By Paul Derrington, Jr.
CEO | Chief Engineer | Research Engineer | Editor, D-CSIL
4 min read
Key Takeaways
Microsoft used Ignite on November 17 to launch seven MAI models for coding, security, images, voice, transcription and routing. Microsoft's pitch is not a bigger frontier model but near-frontier results at a price that lets agents run around the clock, based mostly on tests Microsoft ran itself.

What did Microsoft actually launch on Tuesday?
At its Ignite conference in San Francisco on November 17, Microsoft AI released what it calls a family of seven new models: MAI Cyber 2 Flash for security work, MAI Code 2 and MAI Code 2 Flash for coding, MAI Image 2.7 for image generation and editing, MAI Realtime 1 for live speech-to-speech conversation, Dragon Copilot Transcribe built on MAI Transcribe 2, and Decision 1.1 for routing each request to a model of the right size. Reuters, reporting from Ignite the same day, described the same suite — a coding model, a cyber-specific model, and new image and transcription models — as Microsoft races to challenge rivals such as Anthropic and OpenAI.
Microsoft AI chief Mustafa Suleyman told Reuters the point is building models for specific domains, which keeps costs low while matching rivals. “This isn't just one model. It's not a flash in the pan,” he said. “This is a system, a hill-climbing machine that we have built, which co-optimizes performance and price.” Microsoft also said its second-generation Maia 200 accelerator is now in production running its largest models, at 1.7 times more tokens per watt — a Microsoft figure.
How close is Code 2, really?
Close on Microsoft's chosen tests, with the conditions printed in its own notes. Averaged over four benchmarks aimed at everyday enterprise coding — DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1 — Microsoft says MAI Code 2 reaches 98% of GPT-6.1 Sol's performance at about 40% of the price, and 96% of Claude Opus 5.5's performance at 5% of the price. Code 2 Flash, for high-volume routine work, is put at comparable performance to GPT-6 Luna at 65% of the price and to Claude Haiku 5.5 at 20%. In a blind side-by-side test of front-end work, Microsoft says developers preferred Code 2 over GPT-6 Sol 64% of the time.
Read those numbers as Microsoft's, not as independent results. The coding comparisons are costed at 50% promotional prices, and Microsoft's launch posts give no list price for either Code 2 model. Three of the four coding benchmarks ran on MSBench, Microsoft's internal evaluation platform, and Microsoft itself notes that comparisons vary by workload, configuration, reasoning effort and pricing assumptions. The models are also, in Microsoft's words, explicitly not yet tuned for harder problems such as fundamental science research and very long-horizon agentic work.
What do the security, image and voice models claim?
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 with both models costed at list prices — $0.10 per million input tokens and $0.50 per million output tokens for Cyber 2 Flash in Microsoft's notes. The model takes up to 1 million tokens at a time and is priced 86% lower on output tokens than MAI Cyber 1.1 Flash, Microsoft says. One disclosure matters: 92 of the benchmark's 905 tasks draw on the same projects and vulnerability reports as some of its training data, so Microsoft excluded those tasks for every model and reports results on the remaining 813.
For images, Microsoft says MAI Image 2.7 ranks first on the Artificial Analysis image-editing leaderboard as of November 16, at about $35 per 1,000 images in Foundry public preview, with the faster Image 2.7 Flash at about $10 per 1,000. MAI Realtime 1, Microsoft's first live speech-to-speech model, starts at $0.01 per audio minute, with reasoning billed separately. Dragon Copilot Transcribe is live in Dragon Copilot and, Microsoft says, handles 10 million patient encounters a month with multilingual error rates halved. Decision 1.1, built on Gemma and now generally available, is put by Microsoft at 50 times faster and 79 times cheaper than GPT-Sol 6.1, used by 4,000 customers.
Who can use what, starting when?
Availability differs by model. Code 2 and Code 2 Flash can be tried today in GitHub Copilot; Code 2 Flash is also in a limited-access preview in Microsoft Foundry. Image 2.7 and Image 2.7 Flash are in public preview in Foundry, and Decision 1.1 is generally available in the Foundry catalog, now with image inputs. Realtime 1 can be tried in MAI Labs, with a form to register interest.
Cyber 2 Flash is the narrowest launch: Microsoft says it is coming exclusively to Microsoft security products, including Codename MDASH, where it will automatically replace the existing model in the MAI profile for eligible customers from mid-December. The other new offer, MAI Licensed Weights — take a frontier MAI model, customize it with your own data, and run it in your own Azure environment — is described as coming soon through Frontier Company engagements, not as available today.