Moonshot AIOpen weightsKimi K2 0905
Kimi K2 0905 is a Moonshot AI open-weights model, released 5 Sept 2025. Its best configuration ranks #280 of 372 on the BenchLeader Index at 51.0 ±4.4, in the lower half. It scores highest in human preference (61) and lowest in instruction following (44). At $1.07 per million tokens blended it is mid-priced. Output speed of 40 tokens per second puts it in the slowest quarter, with a first answer in 1.2 s. Last measured 2 Sept 2026.
- Blended price
- $1.07/M
- $0.600 in · $2.50 out
- Output speed
- 40 tok/s
- measured by Artificial Analysis
- First answer
- 1.19 s
- first token 0.66 s
- Context
- 262k
- Overall index51
- Reasoning55
- Coding51
- Agents & tools54
- Knowledge52
- Instruction following44
- Human preference61
- Long context46
- Composite48
Versions
Moonshot AI has shipped 3 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| Kimi K2thinking | 6 Nov 2025 | 55.2 | #186 |
| Kimi K2 0905this page | 5 Sept 2025 | 51.0 | #280 |
| Kimi K2 | 11 Jul 2025 | 49.1 | #322 |
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 | default | Source |
|---|---|---|
| LMArena Hard Prompts | 1436#110 | LMArena |
| GPQA Diamond (AA)not in index | 76.7%#226 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 6.4%#328 | Artificial Analysis |
| Kagi LLM Benchmark | 52.2%#76 | Kagi LLM Benchmark |
Coding
| Benchmark | default | Source |
|---|---|---|
| WeirdML | 36.7%#122 | WeirdML |
| LMArena Coding | 1467#103 | LMArena |
| SWE-bench Verified (any scaffold)not in index | 71.2%#14 | SWE-bench |
Agents & tools
| Benchmark | default | Source |
|---|---|---|
| Terminal-Bench Hard | 23.5%#144 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 73.4%#141 | Artificial Analysis |
Knowledge
| Benchmark | default | Source |
|---|---|---|
| AA-Omniscience | -26.6#207 | Artificial Analysis |
Instruction following
| Benchmark | default | Source |
|---|---|---|
| IFBench | 41.7%#240 | Artificial Analysis |
Human preference
| Benchmark | default | Source |
|---|---|---|
| LMArena Text | 1418#108 | LMArena |
Long context
| Benchmark | default | Source |
|---|---|---|
| Fiction.LiveBench 120k | 40.6%#25 | Fiction.live |
| AA-LCR | 53.0%#247 | Artificial Analysis |
Composite
| Benchmark | default | Source |
|---|---|---|
| AA Intelligence Index | 15.3#223 | 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 | 32 tok/s | 0.66 s | $0.600 | $2.50 | 262k | 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 | – | 8.7 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0087 | – | 16.3 s |
| Code edit | 6,000 / 1,500 | $0.0074 | – | 38.9 s |
| Agentic coding session | 60,000 / 4,000 | $0.046 | – | 1.7 min |
| Structured extraction | 2,000 / 200 | $0.0017 | – | 6.2 s |
See also
Data as of 10 Sept 2026. Compare with another model.