Kimi-K2-Instruct
Language model from Moonshot, 1T parameters.
Moonshot · Text line
5 stages over 12 months, from Kimi-K2-Instruct in Jul 2025 to Kimi K3 in Jul 2026.
Language model from Moonshot, 1T parameters.
Every generation. A new possibility.
Kimi-K2-Instruct · Jul 20255 stages, from Kimi-K2-Instruct (Jul 2025) to Kimi K3 (Jul 2026). Press Evolve to watch each step.
Context window
Max output
Output priceper 1M tokens, lower is better
Parameters
Downloadslast 30 days
v2 Stage 1
Language model from Moonshot, 1T parameters.
v2.5 Stage 2
Vision-language model from Moonshot, 1T parameters.
v2.6 Stage 3
Multimodal Kimi workhorse for agent loops, coding tasks, and visual context
v2.7 Stage 4
Coding-focused Kimi model, stronger on long-horizon repo work with less overthinking
v3 Stage 5
Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work