Moonshot AIKimi K3
Best configuration ranks #18 of 610 on the BenchLeader Index at 66.5 ±3.4. Last measured 5 Sept 2026. Released 16 Jul 2026.
- Blended price
- $6.00/M
- $3.00 in · $15.00 out
- Output speed
- 37 tok/s
- First answer
- 58 s
- first token 1.84 s
- Context
- 1.0M
- Overall index67
- Coding65
- Agents & tools68
- Maths78
- Knowledge68
- Human preference71
- Multimodal65
- Long context71
- Composite74
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 |
|---|---|---|---|---|---|---|
| low | 58.8 | #117 | 37 tok/s | 58 s | $0.0057 | Coding 59 · Composite 72 · Knowledge 66 · Long context 66 · Maths 54 · Multimodal 63 · Reasoning 51 |
| high | – | – | 37 tok/s | 58 s | – | Maths 65 · Reasoning 63 |
| max | 62.7 | #51 | 37 tok/s | 58 s | – | Agents & tools 63 · Coding 70 · Human preference 69 · Knowledge 60 · Maths 63 · Reasoning 64 |
| thinking | – | – | 37 tok/s | 58 s | – | Maths 65 |
| defaultbest | 66.5 | #18 | 37 tok/s | 58 s | $0.0057 | Agents & tools 68 · Coding 65 · Composite 74 · Human preference 71 · Knowledge 68 · Long context 71 · Maths 78 · Multimodal 65 |
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 | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| GPQA Diamond | 84.8%#81 | 91.9%#22 | 93.1%#17 | – | – | Epoch AI Benchmarking Hub | |
| SimpleBench | – | – | 60.7%#28 | – | – | SimpleBench | |
| LMArena Hard Prompts | – | – | 1519#7 | – | – | LMArena | |
| LiveBench Reasoningnot in index | – | – | – | – | 90.7%#5 | LiveBench | |
| GPQA Diamond (AA)not in index | 84.2%#144 | – | – | – | 93.5%#14 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 25.0%#135 | – | – | – | 46.9%#24 | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | – | – | – | – | 92.9%#12 | Vals AI | |
| ARC-AGI-1 | 65.7%#101 | 86.7%#66 | 94.5%#24 | – | – | ARC Prize | |
| ARC-AGI-2 | 12.4%#103 | 55.0%#66 | 60.4%#56 | – | – | ARC Prize |
Coding
| Benchmark | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| SciCode | 51.2%#53 | – | 58.7%#5 | – | 58.7%#5 | SciCode | |
| WeirdML | – | – | 82.6%#15 | – | – | WeirdML | |
| FrontierCode | – | – | – | – | 44.2%#8 | Cognition | |
| LMArena Coding | – | – | 1542#6 | – | – | LMArena | |
| LMArena WebDev | – | – | 1674#5 | – | – | LMArena | |
| LiveBench Codingnot in index | – | – | – | – | 81.5%#9 | LiveBench | |
| SciCode (AA)not in index | 52.7%#51 | – | – | – | 59.5%#6 | Artificial Analysis | |
| LiveCodeBench | – | – | 87.2%#17 | – | – | Vals AI | |
| SWE-bench (Vals)not in index | – | – | – | – | 93.4%#8 | Vals AI |
Agents & tools
| Benchmark | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| APEX-Agents | – | – | – | – | 39.3%#11 | Mercor | |
| LMArena Agent | – | – | 6.6#8 | – | – | LMArena | |
| LiveBench Agentic Codingnot in index | – | – | – | – | 62.2%#6 | LiveBench | |
| Terminal-Bench 2.1 (Vals) | – | – | – | – | 80.9%#6 | Vals AI | |
| MCP Atlas | – | – | 82.3%#6 | – | – | Scale AI SEAL |
Maths
| Benchmark | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| FrontierMath Tiers 1–3 | – | – | 72.2%#15 | – | – | Epoch AI Benchmarking Hub | |
| FrontierMath Tier 4 | – | – | 39.0%#22 | – | – | Epoch AI Benchmarking Hub | |
| OTIS Mock AIME | 68.9%#126 | 93.3%#44 | 97.2%#23 | – | – | Epoch AI Benchmarking Hub | |
| ProofBench | – | – | – | – | 87.0%#5 | Vals AI | |
| LiveBench Mathematicsnot in index | – | – | – | – | 84.4%#41 | LiveBench | |
| AIME 2026 | – | – | – | 96.7%#6 | – | MathArena | |
| HMMT February 2026 | – | – | – | 97.0%#3 | – | MathArena | |
| MathArena Apex | – | – | – | 65.6%#4 | – | MathArena |
Knowledge
| Benchmark | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| SimpleQA Verified | – | – | 50.6%#20 | – | – | Epoch AI Benchmarking Hub | |
| LiveBench Data Analysisnot in index | – | – | – | – | 78.7%#14 | LiveBench | |
| AA-Omniscience | 3.9#77 | – | – | – | 19.7#43 | Artificial Analysis | |
| MMLU-Pro | – | – | 88.0%#22 | – | – | Vals AI | |
| LegalBench | – | – | – | – | 86.0%#14 | Vals AI | |
| CorpFin | – | – | – | – | 71.6%#3 | Vals AI | |
| TaxEval | – | – | – | – | 75.7%#12 | Vals AI |
Instruction following
| Benchmark | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| LiveBench Languagenot in index | – | – | – | – | 85.5%#8 | LiveBench |
Human preference
| Benchmark | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| LMArena Text | – | – | 1489#12 | – | – | LMArena | |
| EQ-Bench 4 | – | – | – | – | 1339#3 | EQ-Bench |
Multimodal
| Benchmark | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| MMMU-Pro | 78.5%#52 | – | – | – | 80.5%#31 | Artificial Analysis |
Long context
| Benchmark | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| AA-LCR | 79.3%#65 | – | – | – | 88.7%#1 | Artificial Analysis |
Composite
| Benchmark | low | high | max | thinking | default | Source | Trend |
|---|---|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | – | – | 157.6#11 | Epoch AI Benchmarking Hub | |
| LiveBench | – | – | – | – | 79.2%#8 | LiveBench | |
| AA Intelligence Index | 34.5#57 | – | – | – | 43.8#24 | Artificial Analysis | |
| Vals Indexnot in index | – | – | – | – | 57.8#15 | Vals AI |
Where it wins
Benchmarks where this configuration ranks in the top five of every configuration measured.
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 |
|---|---|---|---|---|---|---|
| Modal | 80 tok/s | 0.99 s | $3.00 | $15.00 | 1.0M | mxfp4 |
| Relace | 68 tok/s | 1.52 s | $2.40 | $12.00 | 1.0M | fp4 |
| Together | 68 tok/s | 1.28 s | $3.00 | $15.00 | 1.0M | – |
| Sail Research | 57 tok/s | 1.68 s | $2.60 | $13.00 | 1.0M | fp4 |
| Parasail | 55 tok/s | 1.37 s | $3.00 | $15.00 | 1.0M | fp4 |
| Makora | 54 tok/s | 1.43 s | $2.55 | $12.75 | 1.0M | – |
| Wafer | 50 tok/s | 0.74 s | $3.00 | $12.75 | 1.0M | – |
| Fireworks (US) | 49 tok/s | 2.24 s | $3.30 | $16.50 | 1.0M | – |
| Fireworks | 44 tok/s | 2.16 s | $3.00 | $15.00 | 1.0M | – |
| Fireworks Fast | 40 tok/s | 1.13 s | $4.50 | $22.50 | 1.0M | – |
| Baseten | 36 tok/s | 1.80 s | $3.00 | $15.00 | 1.0M | fp8 |
| Moonshot AI | 26 tok/s | 3.90 s | $3.00 | $15.00 | 1.0M | mxfp4 |
| Alibaba Cloud Int. | 25 tok/s | 1.84 s | $3.45 | $17.25 | 1.0M | – |
| DigitalOcean | 24 tok/s | 2.52 s | $2.85 | $14.25 | 1.0M | – |
| Phala | 22 tok/s | 4.12 s | $3.00 | $15.00 | 1.0M | – |
| Chutes | 21 tok/s | 2.54 s | $3.00 | $15.00 | 1.0M | mxfp4 |
| Morph Fast | 13 tok/s | 3.24 s | $6.00 | $22.50 | 1.0M | fp4 |
| Morph | 12 tok/s | 3.45 s | $2.80 | $14.00 | 1.0M | fp4 |
| DeepInfra | 12 tok/s | 3.70 s | $2.85 | $14.25 | 1.0M | bf16 |
Price history
Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.
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 $0.300 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0057 | $0.0049 | 1.1 min |
| Summarise a 30-page report | 12,000 / 600 | $0.045 | $0.021 | 1.2 min |
| Code edit | 6,000 / 1,500 | $0.041 | $0.028 | 1.6 min |
| Agentic coding session | 60,000 / 4,000 | $0.240 | $0.118 | 2.8 min |
| Structured extraction | 2,000 / 200 | $0.0090 | $0.0050 | 1.1 min |
See also
Data as of 9 Sept 2026. Compare these configurations.