Moonshot AIKimi K2
Best configuration ranks #313 of 610 on the BenchLeader Index at 48.8 ±3.3. Last measured 2 Sept 2026. Released 11 Jul 2025.
- Blended price
- $1.07/M
- $0.600 in · $2.50 out
- Output speed
- 40 tok/s
- First answer
- 1.48 s
- first token 0.94 s
- Context
- 262k
- Overall index49
- Reasoning43
- Coding47
- Agents & tools54
- Maths52
- Knowledge45
- Instruction following44
- Human preference60
- Long context46
- Composite48
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| thinking | – | – | 40 tok/s | 1.48 s | $0.0009 | Reasoning 59 |
| defaultbest | 48.8 | #313 | 40 tok/s | 1.48 s | $0.0010 | Agents & tools 54 · Coding 47 · Composite 48 · Human preference 60 · Instruction following 44 · Knowledge 45 · Long context 46 · Maths 52 · Reasoning 43 |
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | thinking | default | Source |
|---|---|---|---|
| SimpleBench | – | 26.3%#71 | SimpleBench |
| LMArena Hard Prompts | – | 1436#111 | LMArena |
| GPQA Diamond (AA)not in index | – | 76.7%#223 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | – | 7.4%#294 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | – | 71.5%#88 | Vals AI |
| Kagi LLM Benchmark | 64.4%#39 | 52.2%#76 | Kagi LLM Benchmark |
Coding
| Benchmark | thinking | default | Source |
|---|---|---|---|
| WeirdML | – | 39.4%#106 | WeirdML |
| GSO-Bench | – | 4.9%#21 | GSO-Bench |
| LMArena Coding | – | 1467#104 | LMArena |
| LiveCodeBench | – | 70.5%#85 | Vals AI |
| IOI | – | 1.3%#56 | Vals AI |
| SWE-Bench Pro | – | 27.7%#17 | Scale AI SEAL |
| Aider Polyglot | – | 59.1%#17 | Aider polyglot leaderboard |
| SWE-bench Verified (bash only) | – | 43.8%#34 | SWE-bench |
| SWE-bench Verified (any scaffold)not in index | – | 71.2%#14 | SWE-bench |
Agents & tools
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Terminal-Bench | – | 27.8%#46 | Terminal-Bench |
| Terminal-Bench Hard | – | 23.5%#143 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | – | 73.4%#140 | Artificial Analysis |
| BFCL Overall | – | 59.1%#9 | Berkeley Function Calling Leaderboard |
Maths
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AIME (Vals) | – | 62.7%#55 | Vals AI |
| MGSM | – | 91.0%#30 | Vals AI |
Knowledge
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AA-Omniscience | – | -26.6#203 | Artificial Analysis |
| MMLU-Pro | – | 79.4%#93 | Vals AI |
| LegalBench | – | 81.5%#74 | Vals AI |
| CorpFin | – | 50.4%#103 | Vals AI |
| TaxEval | – | 70.2%#90 | Vals AI |
| MedQA | – | 84.0%#60 | Vals AI |
| PRBench Finance | – | 38.3%#28 | Scale AI SEAL |
| PRBench Legal | – | 36.4%#30 | Scale AI SEAL |
| MultiNRC | – | 18.5%#33 | Scale AI SEAL |
Instruction following
| Benchmark | thinking | default | Source |
|---|---|---|---|
| IFBench | – | 41.7%#240 | Artificial Analysis |
Human preference
| Benchmark | thinking | default | Source |
|---|---|---|---|
| LMArena Text | – | 1418#108 | LMArena |
Long context
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Fiction.LiveBench 120k | – | 40.6%#25 | Fiction.live |
| AA-LCR | – | 53.0%#243 | Artificial Analysis |
Composite
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | 140.1#98 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | – | 15.3#218 | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| NovitaAI | 19 tok/s | 0.94 s | $0.570 | $2.30 | 131k | fp8 |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0010 | – | 8.9 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0087 | – | 16.4 s |
| Code edit | 6,000 / 1,500 | $0.0074 | – | 38.7 s |
| Agentic coding session | 60,000 / 4,000 | $0.046 | – | 1.7 min |
| Structured extraction | 2,000 / 200 | $0.0017 | – | 6.4 s |
See also
Data as of 9 Sept 2026. Compare these configurations.