Moonshot AIKimi K2 Thinking
Best configuration ranks #238 of 610 on the BenchLeader Index at 52.8 ±3.7 (thinking reasoning effort). Last measured 1 Sept 2026. Released 6 Nov 2025.
- Blended price
- $1.07/M
- $0.600 in · $2.50 out
- Output speed
- 119 tok/s
- First answer
- 18 s
- first token 0.93 s
- Context
- 262k
- Overall index53
- Coding49
- Agents & tools44
- Maths50
- Knowledge52
- Instruction following58
- Long context62
- Composite57
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 | Source |
|---|---|---|
| GPQA Diamond (AA)not in index | 83.8%#151 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 23.8%#140 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | 78.5%#70 | Vals AI |
Coding
| Benchmark | thinking | Source |
|---|---|---|
| WeirdML | 42.8%#90 | WeirdML |
| LiveCodeBench | 63.1%#98 | Vals AI |
| SWE-bench (Vals)not in index | 60.2%#74 | Vals AI |
| SWE-bench Verified (bash only) | 63.4%#21 | SWE-bench |
| SWE-bench Verified (any scaffold)not in index | 63.4%#26 | SWE-bench |
Agents & tools
| Benchmark | thinking | Source |
|---|---|---|
| Terminal-Bench | 35.7%#40 | Terminal-Bench |
| APEX-Agents | 4.1%#58 | Mercor |
| Terminal-Bench Hard | 31.1%#104 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 93.0%#47 | Artificial Analysis |
Maths
| Benchmark | thinking | Source |
|---|---|---|
| AIME (Vals) | 85.4%#36 | Vals AI |
| MGSM | 90.2%#39 | Vals AI |
| MathArena Apex | 0.0%#46 | MathArena |
Knowledge
| Benchmark | thinking | Source |
|---|---|---|
| AA-Omniscience | -21.4#185 | Artificial Analysis |
| MMLU-Pro | 81.1%#81 | Vals AI |
| LegalBench | 80.2%#81 | Vals AI |
| CorpFin | 60.6%#67 | Vals AI |
| TaxEval | 71.7%#72 | Vals AI |
| MedQA | 92.6%#28 | Vals AI |
| PRBench Finance | 43.4%#20 | Scale AI SEAL |
| PRBench Legal | 40.9%#21 | Scale AI SEAL |
Instruction following
| Benchmark | thinking | Source |
|---|---|---|
| IFBench | 68.1%#81 | Artificial Analysis |
| MultiChallenge | 55.4%#18 | Scale AI SEAL |
Long context
| Benchmark | thinking | Source |
|---|---|---|
| AA-LCR | 72.0%#140 | Artificial Analysis |
Composite
| Benchmark | thinking | Source |
|---|---|---|
| Epoch Capabilities Indexnot in index | 145.8#67 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 22.0#152 | 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 |
|---|---|---|---|---|---|---|
| Google Vertex | 159 tok/s | 0.72 s | $0.600 | $2.50 | 262k | – |
| NovitaAI | 29 tok/s | 1.15 s | $0.600 | $2.50 | 262k | bf16 |
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 at $0.150 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0010 | $0.0009 | 20.8 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0087 | $0.0046 | 23.3 s |
| Code edit | 6,000 / 1,500 | $0.0074 | $0.0053 | 30.9 s |
| Agentic coding session | 60,000 / 4,000 | $0.046 | $0.026 | 52.0 s |
| Structured extraction | 2,000 / 200 | $0.0017 | $0.0010 | 20.0 s |
See also
Data as of 9 Sept 2026. Compare with another model.