[X]score: 0.77
Microsoft MAI-Thinking-1: 35B MoE, 97% AIME, 53% SWE-Bench Pro
June 2, 2026
Microsoft launched seven MAI models headlined by MAI-Thinking-1, a 35B active parameter MoE with 256K context, scoring 97% on AIME 2025 and 53% on SWE-Bench Pro, competitive with Claude Opus 4.6. Running on Microsoft's MAIA 200 chip, it delivers 30% better performance per dollar and 1.4x performance-per-watt versus Nvidia GB200.
HOW THIS AFFECTS YOU
●
builderYou now have a 35B MoE reasoning model with 256K context optimized for coding tasks at potentially lower inference cost than comparable Anthropic models.
●
researcherWorth watching because the MAIA 200 co-design approach yields measurable efficiency gains over GB200, suggesting hardware-model co-optimization is a viable path to cost reduction.
●
founderA credible third frontier coding model at competitive benchmarks increases your negotiating leverage and reduces dependency on Anthropic or OpenAI.
●
investorMicrosoft vertically integrating silicon and models directly challenges Anthropic and OpenAI on cost structure, signaling a major competitive shift in the frontier model market.