Chinese AI startup MiniMax has launched its new flagship model, M3, which is engineered for coding tasks and automated workflows. This model boasts a significantly reduced computational need, requiring only one-twentieth of previous levels, and can process up to one million tokens simultaneously. MiniMax asserts that M3 outperforms GPT-5.5 and Gemini 3.1 Pro on coding benchmarks. AI
IMPACT Sets new SOTA on coding benchmarks; pressures OpenAI and Google to respond.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Mastodon — sigmoid.social →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →