Moonshot AI has launched Kimi K3, a new open-weights frontier-class model boasting 2.8 trillion parameters and a 1 million token context window. The model features native multimodal input capabilities and a novel Kimi Delta Attention mechanism for faster decoding. While Kimi K3 has achieved top rankings in coding benchmarks, surpassing models like Claude Fable 5 and GPT-5.6 Sol in specific arenas, its developers acknowledge a gap in user experience compared to these leading closed models. AI
IMPACT Sets a new benchmark for open-source models in coding tasks, potentially accelerating development and competition in the AI space.
RANK_REASON Frontier-lab model release with system card and benchmark results. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- 1M-token
- 2.8T
- Agent Arena
- Claude Fable 5
- Frontend Code Arena
- GPT-5.6 Sol
- Kimi Delta Attention
- Kimi K3
- Moonshot AI
- Opus 4.8
- Sonnet 5
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
