ZhiKe AI

Apr 21, 2026 · Artificial Intelligence

Open-Source Kimi K2.6 Beats GPT‑5.4 and Claude Opus 4.6 in Code Generation

Kimi K2.6, an open‑source Chinese LLM, outperforms GPT‑5.4 and Claude Opus 4.6 on SWE‑Bench Pro code tests, delivers 13‑hour uninterrupted coding, runs 300 parallel agents, and costs only one‑twentieth of comparable closed‑source models, while offering a trillion‑parameter MoE architecture and Apache 2.0 licensing.