**Moonshot Kimi K2.6 1T MoE OSS release + Qwen3.6-Max shootouts coding SOTA**
Key Questions
What is Kimi K2.6?
Kimi K2.6 is Moonshot AI's 1T MoE open-source release on HuggingFace, claiming SOTA on HLE-Full 54, SWE-Pro 58.6, and 20+ benchmarks. It excels in long-horizon coding, tools, vision sketch-to-web, and scales to 300 agents with 4,000 steps. It edges GPT-5.4 and Claude 4.6.
How does Kimi K2.6 compare to competitors?
Kimi K2.6 leads in coding SOTA, MLA efficiency, and Rust tasks over GPT-5.4/Claude 4.6. Qwen3.6-Max-Preview surpasses it in SWE-Pro, TerminalBench2.0, and 6 programming evals. Chinese OSS challenges Western leaders.
What are the strengths of Qwen3.6-Max?
Qwen3.6-Max preview leads programming benchmarks like SWE-Pro and TerminalBench2.0 over Kimi K2.6 and Plus models. It highlights the Chinese OSS surge. Shootouts confirm its coding dominance.
Kimi K2.6 OSS HF claims SOTA HLE-Full 54/SWE-Pro 58.6/20+ benches long-horizon coding/tools/vision sketch-to-web/MLA efficiency/300 agents claw/Rust; edges GPT-5.4/Claude 4.6; Qwen3.6-Max-Preview leads SWE-Pro/TerminalBench2.0/6 programming > K2.6/Plus; Chinese OSS surge challenging leaders.