Microsoft unveils six specialized MAI models with lower token pricing
Microsoft introduced six task‑specific MAI models at Ignite, claiming near‑parity with premium rivals at a fraction of the cost, with promotional pricing set to double in 2027.
Microsoft introduced six task‑specific MAI models at Ignite, claiming near‑parity with premium rivals at a fraction of the cost, with promotional pricing set to double in 2027.
At the Ignite conference on November 17, Microsoft announced six new task‑oriented models under its Microsoft AI (MAI) portfolio. The lineup—Code 2, Code 2 Flash, Cyber 2 Flash, Image 2.7, Transcribe 2 and Realtime 1—targets specific workloads such as software development, cybersecurity, image editing, transcription and real‑time interaction. Microsoft says the models deliver performance comparable to leading proprietary systems while costing substantially less, and they are now available through GitHub Copilot and Azure.
Microsoft highlighted Code 2’s results on four coding benchmarks (DeepSWE 1.1, Terminal‑Bench 2.1, SWE Atlas and FrontierCode 1.1). According to the company, Code 2 reached 96% of Claude Opus 5.5’s performance while costing only 5% per task, and 98% of GPT‑6.1 Sol’s performance at 40% of the cost. For cybersecurity, Cyber 2 Flash posted a 92.3% score on the CyberGym‑E2E benchmark, edging out GPT‑6 Astra’s 91.5% score. The company cautioned that these figures are based on internally selected benchmarks and have not yet been independently verified.
The MAI Code 2 model is priced at $0.50 per million input tokens and $2.50 per million output tokens, while the flash variant is offered at $0.05/$0.25 respectively. By contrast, Claude Opus 5.5 charges $4/$20 and GPT‑6.1 Sol $2/$10 for the same units. Microsoft noted that these rates are promotional and will double after May 17, 2027. The lower‑cost flash models are positioned as entry points for developers, with enterprise customers also able to license model weights for custom deployments, though such custom models must run on Microsoft’s Azure cloud.
The launch aligns with Microsoft’s broader strategy to reduce reliance on external AI providers, a concern raised by CEO Satya Nadella earlier in the year. By bundling cheaper models with Azure, Microsoft could increase cloud lock‑in for businesses, a point noted by an Axios interview with AI chief Mustafa Suleyman. Analysts note that while token prices across the industry have fallen sharply since 2023, leading providers like Anthropic and OpenAI still dominate enterprise spend, according to Ramp’s October 2026 index. The concentration of AI chips in Taiwan’s TSMC and data centers in the United States adds regulatory scrutiny, as highlighted in a recent OECD report on market competition.
The next milestone will be the price adjustment scheduled for May 2027, which will test whether customers remain on Azure despite higher costs. Independent benchmark studies are also expected to emerge, providing external validation of Microsoft’s performance claims. Finally, the rollout of customized MAI models may prompt regulatory attention around cloud‑provider lock‑in and competition in the AI infrastructure market.
The introductory rates for MAI Code 2 and Code 2 Flash will double after May 17, 2027.
Custom‑trained MAI models must be hosted on Microsoft’s Azure cloud, though the base models are accessible via GitHub Copilot.
Microsoft reports Code 2 achieves 96% of Claude Opus 5.5’s benchmark performance at 5% of the cost and 98% of GPT‑6.1 Sol’s performance at 40% of the cost.
Crypto news with a focus on Asia: Hong Kong, Singapore, Korea and Japan.
Discussion 0