Output tokens cost $2.19 per million on Kimi K2 Thinking and $15.00 per million on Kimi K3. Input tokens cost $0.55 per million on Kimi K2 Thinking and $3.00 per million on Kimi K3. Cached input tokens are billed at $0.14 per million on Kimi K2 Thinking and $0.30 per million on Kimi K3. Context lengths are 262,144 tokens on Kimi K2 Thinking and 1,048,576 tokens on Kimi K3. Time to first token (TTFT) measured on AIHubMix is 0.4s on Kimi K2 Thinking and 0.9s on Kimi K3. Measured output throughput is 208.3 tok/s on Kimi K2 Thinking and 73.1 tok/s on Kimi K3. On the LMArena coding leaderboard Kimi K2 Thinking scores 1467 and Kimi K3 scores 1531.
Kimi K2 ThinkingKimi K3Kimi K2 Thinking is Moonshot AI's most advanced open-source inference model to date, extending the K2 series into intelligent agent and long-context inference domains. The model is built on the trillion-parameter mixture of experts (MoE) architecture introduced by Kimi K2, activating 32 billion parameters per forward pass and supporting a context window of 256,000 tokens.
Kimi K3 is Kimi’s flagship model for long-horizon coding and end-to-end knowledge work, with a 1M-token context window and industry-leading intelligence.
Prices are per million tokens. Time to First Token and throughput are rolling averages measured on AIHubMix.
Promotional prices show the discounted rate; see each model page for promotion windows.
Daily traffic served through AIHubMix — how demand for each model is trending.
Measured on real AIHubMix traffic, hourly buckets. Gaps mean no traffic in that hour.
LMArena ratings by capability (Bradley-Terry, commonly called Elo). Higher is better.
Source: LMArena (arena.ai) leaderboard, imported by AIHubMix. Models without published ratings are omitted per chart.
Estimate your monthly bill for the same workload on each model.
Monthly = daily × 30. Discounted rates applied where a promotion is active.
Kimi K2 Thinking: $2.19/M output tokens; Kimi K3: $15.00/M. Use the cost calculator above to estimate your own workload.
Kimi K3: 1531; Kimi K2 Thinking: 1467 (LMArena coding leaderboard).
Kimi K2 Thinking: 0.4s time to first token measured on AIHubMix; see the live performance charts above for how each model behaves across the day.
Kimi K2 Thinking accepts 262,144 and Kimi K3 accepts 1,048,576 input tokens. Maximum output per request is 262,144 tokens on Kimi K2 Thinking and 1,048,576 tokens on Kimi K3.
Kimi K2 Thinking at 208.3 tok/s and Kimi K3 at 73.1 tok/s, measured as output throughput on AIHubMix — a separate metric from time to first token.
Kimi K2 Thinking accepts text input and supports thinking, tool calling, function calling and structured outputs; Kimi K3 accepts text, image and video input and supports thinking, function calling and structured outputs.
Yes. AIHubMix serves every model on this page behind one OpenAI-compatible endpoint, so switching between them is a one-line change to the model field — no second account, key or SDK.
Related model match-ups readers also look at.
AIHubMix