Seven new models at scalable price-performance
Points and comments are a snapshot, not live.
Microsoft released seven specialized AI models at dramatically lower prices than competitors.
At Ignite 2026, Microsoft released seven specialized models under MAI. MAI Cyber 2 Flash beats GPT-6 Astra at under 2% of the cost on CyberGym-E2E. MAI Code 2 reaches 98% of GPT-6.1 Sol coding quality at ~40% the price. MAI Image 2.7 tops Artificial Analysis image editing at one-fifth the price of GPT Image 2.6. MAI Realtime 1 outperforms GPT-Realtime 2.1 starting at $0.01/audio minute. Dragon Copilot Transcribe (based on MAI Transcribe 2) powers 10 million patient encounters monthly. Decision 1.1 claims 50x speed and 79x cost improvement over GPT-6.1 Sol. The Maia 200 AI accelerator is in production with 1.7x more tokens per watt.
What commenters are saying
Commenters were sharply divided. Many criticized Microsoft's comparison of Decision 1.1 to GPT-6.1 Sol (a general model, not a decision model) as misleading or useless, with some noting it targets executives over developers. A product team member defended the benchmarks, citing independent leaderboards. A strong camp argued closed-weight models are unappealing; they want open weights for ownership and risk mitigation. Others countered that enterprises prioritize lower cost and performance over openness, and that the market has room for cheaper frontier-competitive API models.