Kimi K2 Thinking
Moonshot AI · released Nov 6, 2025
Kimi K2 Thinking is Moonshot AI’s most advanced open reasoning model to date, extending the K2 series into agentic, long-horizon reasoning. Built on the trillion-parameter Mixture-of-Experts (MoE) architecture introduced in...
Specification
- Context window
- 262K
- Max output
- 100K
- Knowledge cutoff
- Not stated
- Parameters
- 1058.1B
- Licence
- other
- Serving providers
- 2
- Moderated
- No
- Uptime
- 100.0%
Intelligence
33.5
79th percentile
Coding
—
Coding Index
Agentic
—
Agentic Index
Output speed
158 t/s
Median across providers
Latency
429ms
Time to first token
Cost per task
$0.23
Estimated
Benchmarks
Where the score comes from
The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.
Evaluation scores
Percentage correct · higher is better
- AIME 202594.7%
- τ²-bench (Telecom)93.0%
- LiveCodeBench85.3%
- MMLU-Pro84.8%
- GPQA Diamond83.8%
- AA-LCR (long context)70.3%
- IFBench68.1%
- SciCode42.4%
- Terminal-Bench Hard31.1%
- Humanity's Last Exam23.8%
An evaluation missing from this list was not run for this model — it is not a zero.
View as table
| Evaluation | Score |
|---|---|
| AIME 2025 | 94.7% |
| τ²-bench (Telecom) | 93.0% |
| LiveCodeBench | 85.3% |
| MMLU-Pro | 84.8% |
| GPQA Diamond | 83.8% |
| AA-LCR (long context) | 70.3% |
| IFBench | 68.1% |
| SciCode | 42.4% |
| Terminal-Bench Hard | 31.1% |
| Humanity's Last Exam | 23.8% |
Against its peers
Intelligence Index · this model highlighted, nearest peers in grey
- M2.5MiniMax34.5
- Hy3 previewTencent34.4
- 34.3
- LongCat 2.0Meituan33.9
- KAT-Coder-Pro V2KwaiPilot33.9
- Kimi K2 ThinkingMoonshot AI33.5
- o3 ProOpenAI33.3
- 32.8
Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.
View as table
| Model | Intelligence |
|---|---|
| M2.5 | 34.5 |
| Hy3 preview | 34.4 |
| Qwen3.5 397B A17B | 34.3 |
| LongCat 2.0 | 33.9 |
| KAT-Coder-Pro V2 | 33.9 |
| Kimi K2 Thinking | 33.5 |
| o3 Pro | 33.3 |
| Qwen3.5-122B-A10B | 32.8 |
Percentile among all indexed models
Pricing
What it costs to run
List prices per million tokens, plus what one representative task works out to.
List price
- Input / 1M tokens
- $0.6
- Output / 1M tokens
- $2.5
- Cached input / 1M
- $0.15
- Blended 3:1
- $1.07
One task, estimated
$0.23
- Input tokens
- 50,000
- Output tokens
- 80,000
- Profile
- Reasoning
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).
Arena
Head-to-head generation quality
Elo from pairwise judgements, broken out by the kind of thing the model was asked to build.
- Overall Elo
- 1,135
- Win rate
- 48.8%
- Strongest at
- Websites
- Tournaments
- 713
Elo by category
Dot position on a 1,125–1,150 scale · Elo has no meaningful zero
- 1,135
Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.
View as table
| Category | Elo | Win rate |
|---|---|---|
| Websites | 1135 | 48.8% |