MAI Cyber 2 Flash beats GPT-6 Astra quality at less than 2% of the cost per task on CyberGym-E2E, Microsoft AI said at Ignite.
That is a pricing claim, not a capability leap.
MAI Code 2 reaches 98% of the coding quality of GPT-6.1 Sol at about 40% of the price, the company said. Against Opus 5.5 it reaches 96% performance for just 5% of the price, and Microsoft positions both Code models for everyday enterprise coding work.
MAI Code 2 Flash matches GPT-6 Luna at 65% of the price and Haiku 5.5 at 20%, Microsoft said. MAI Image 2.7 ranks first on Artificial Analysis image editing and undercuts GPT Image 2.6 Sunburst (max) at one-fifth of the price.
Microsoft AI also listed MAI Realtime 1 from $0.01 per audio minute, Transcribe 2 powering 10 million patient encounters a month, Decision 1.1 at 50x faster and 79x cheaper than GPT-6.1 Sol, and Maia 200 at 1.7x more tokens per watt, the company said.
The numbers do not show evaluation setup, token counts or sustained throughput. Independent cost-per-task accounting would change the read.
Will buyers trade that last 2% to 4% of quality for a 20x cut in agent run cost?





