Thursday, November 19, 2026 Updated Briefings Saved 中文
AI Daily Radar

AI and the frontier tech around it, explained every morning

Announced

Microsoft's MAI unveils seven models as products reportedly shift in-house

Photo: Hay Kranen / Wikimedia Commons · CC BY 4.0, cropped

At Ignite, Microsoft’s MAI unit unveiled seven in-house models covering code, security, images, language, and routing, and Microsoft is reportedly moving some of its own products off external AI models to cheaper MAI alternatives. Microsoft says MAI Code 2 reaches 98% of GPT-6.1 Sol’s performance at 40% of the cost, and the models are designed to run enterprise AI agents continuously. The report does not specify which products are switching, what prices apply, or whether the models are generally available; the performance and cost figures are Microsoft’s own claims.

Verified and unconfirmed

  • Verified: Microsoft announced seven in-house MAI models at Ignite covering code, security, images, language and routing, and its own announcement states that MAI Code 2 reaches 98% of GPT-6.1 Sol’s coding performance at 40% of the cost, while MAI Code 2 Flash matches GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of their cost respectively (Microsoft, as reported by Techzine).
  • Not yet confirmed: The 98%/40% and Flash figures are vendor benchmarks with no independent replication reported, and one outside analysis frames the comparison as being against a deprecated model (OrcaRouter), which would change how the cost-performance claim should be read. The report’s core assertion that Microsoft is replacing external models inside its own products is attributed only to reporting rather than a named shipping commitment, and the announcement gives no per-token pricing, rollout dates, or list of affected products.

Background

MAI is Microsoft’s in-house model unit, and the seven models it showed at Ignite are positioned as cheaper substitutes for the models Microsoft’s own products call from outside suppliers. Microsoft’s announcement frames the shift as a price-performance trade rather than a capability leap — MAI Code 2 is benchmarked against GPT-6.1 Sol instead of being claimed to beat it — and pitches the family for always-on enterprise agents, a workload where per-task cost compounds quickly.

Why it matters

Enterprise buyers should expect Microsoft to position MAI Code 2 as a cheaper option inside its products, but until Microsoft names the affected products and prices, the 98%/40% comparison cannot be used for procurement decisions.

What to watch

Watch for Microsoft to publish product-by-product rollout details and per-token pricing for the seven MAI models, and for independent benchmarks testing the 98%-of-GPT-6.1-Sol and 40%-cost claims.

Further reading

Comments

Comments

Comments use a GitHub account and are moderated. Comment rules

More in AI

See all