Kimi K3 release: 2.8T params, 1M context, native multimodal

Melvin Vivas · X post · 2026-07-17 · Open on X

Topics: Industry Trends & Job Market, LLM Fundamentals · Level: intermediate

Summary

Announces that Moonshot's Kimi K3 model has been released and lists its reported specs. It's useful as a trend snapshot of frontier open-weight model scale and new attention variants. The creator also asks where people can access it.

Key points

• Kimi K3 reportedly has more than 2.8 trillion parameters.
• It has a 1 million token context window.
• It is natively multimodal.
• It uses architecture changes called Kimi Delta Attention and Attention Residual.
• Access options were still unclear when it was posted.

Resources mentioned

• Kimi K3 · tool · huggingface.co · free
  Moonshot AI's large multimodal LLM with a 1M-token context window and Kimi Delta Attention.
  Also in: Multi-Teacher On-Policy Distillation (MOPD) in 2026 (Melvin Vivas on X · notes), Qwen3.8-Max on the Frontend Code Arena cost-performance frontier (Melvin Vivas on X · notes), 1-bit Kimi K3 GGUF Running Locally vs Claude Opus 5 and GPT 5.6 (Melvin Vivas on X · notes), Code Arena Fullstack Benchmark: Kimi K3 Ranks #1 (Melvin Vivas on X · notes) and 1 more