MiniMax M3 coding LLM claims top SWE-bench Pro score

A new open-weight coding model from MiniMax, M3, has emerged as a top performer, claiming the leading score on the SWE-bench Pro benchmark. Released on June 1, 2026, M3 boasts impressive claims, including a 1M-token context window and multimodal capabilities, with a novel sparse attention architecture designed for faster processing. However, these benchmarks are self-reported by MiniMax, and independent verification is still pending, with some noting comparisons to slightly older model versions. AI

IMPACT Sets a new benchmark for open-weight coding models, potentially influencing future development and competition.

RANK_REASON New model release from a significant lab with benchmark claims. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• Claude Opus 4.7
• Claude Opus 4.8
• Codestral
• DeepSeek
• Gemini 3.1 Pro
• GPT-5.5
• Llama
• MiniMax
• OpenAI
• Qwen
• SWE-bench Pro

• model release
• paper

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →