AI Research Pulse

MiniMax-M3 Challenges Frontier Models with 5-10% Cost

MiniMax-M3 Challenges Frontier Models with 5-10% Cost

Key Questions

How does MiniMax-M3 compare to frontier models?

MiniMax-M3 beats GPT-5.5 and Gemini 3.1 Pro on SWE-Bench Pro while using only 5-10% of the cost via sparse attention.

Where does MiniMax-M3 fall short?

It trails Opus 4.8 on complex reasoning tasks despite strong cost-performance trade-offs.

What trend does MiniMax-M3 support?

It challenges the closed-model oligopoly by offering efficient open alternatives.

MiniMax-M3 beats GPT-5.5 and Gemini 3.1 Pro on SWE-Bench Pro at 5-10% cost using sparse attention (MSA). Trails Opus 4.8 on complex reasoning but offers compelling cost-performance trade-off. This efficient model challenges the frontier closed-model oligopoly and aligns with the trend of cost-effective alternatives.

Sources (2)
Updated Jun 2, 2026