moonshotai/kimi-k2Kimi K2 is Moonshot’s open-weights AI model, released Sep 5, 2025. It scores 140 on the Epoch Capabilities Index, costs $0.57/M input and $2.30/M output tokens, and handles a 131K-token context window.
Kimi K2 Instruct is a large-scale Mixture-of-Experts (MoE) language model developed by Moonshot AI, featuring 1 trillion total parameters with 32 billion active per forward pass.
Scores on standardized evaluations. Higher is better — and rank shows where Kimi K2 lands among all models tracked on Model Beat.
Graduate-level scientific reasoning
Frontier of human expert knowledge
Broad expert knowledge
Common-sense trick questions
Novel ML problem-solving
Code performance optimization
Contamination-free coding
Scientific research coding
Olympiad-qualifier math
Command-line agentic tasks
Tool-agent-user reliability
Kimi K2 Instruct is a large-scale Mixture-of-Experts (MoE) language model developed by Moonshot AI, featuring 1 trillion total parameters with 32 billion active per forward pass.
Kimi K2 is an AI model developed by Moonshot, part of the Kimi family, released Sep 5, 2025.
Kimi K2 is served from $0.57 per million input tokens and $2.30 per million output tokens (cheapest credible provider on OpenRouter: Novita).
Kimi K2 supports a context window of up to 131K tokens.
On standardized evaluations tracked by Epoch AI, Kimi K2 scores 55.6% on LiveCodeBench, 82.4% on MMLU-Pro, 39.4% on WeirdML.
Kimi K2 is released with open weights (Open weights (restricted use)), so the model can be downloaded and self-hosted.
Benchmark data from Epoch AI (CC BY) and Artificial Analysis; pricing, specs & descriptions from OpenRouter. Data updated Nov 12, 2026.