Kimi K2 Thinking
Unavailable in this round
Language modelOpen source
Kimi K2 Thinking
Moonshot AI
- Released
- November 4, 2025
- Data date
- September 8, 2026
- Benchmark-Meta-Score
- Standard error ± 0.1
- Speed
- 120.4 tokens/sOutput throughput
- Price
- Input · Output / 1M tokens
- Input
- Text, Image
- Output
- Text
What is distinctive about Kimi K2 Thinking is its distinct Thinking checkpoint rather than a response variant of Kimi K2. Moonshot describes 1 trillion total parameters, 32 billion active parameters, and a 262,144-token context window.
Kimi’s official API catalog marks kimi-k2-thinking as discontinued from May 25, 2026. The open weights can still be used locally or through third parties, but the catalog entry does not confirm current first-party API availability.
- Position
- Rank 69 of 91
- Index score
- 0.1 / 100
Methodology and source
The LLM comparison hub's selected-panel index averages rank percentiles from 65 documented comparison cohorts. Benchmark meta-scores are shown in the benchmark section.
Benchmark meta-analysis
Overall and area scores
These Gradually indexes place selected published measurements on a 0 to 100 scale. They are not accuracy percentages and do not guarantee results in your application.
Retrieval dates of the sources used 2026-10-15 - 2026-10-19
Overall index
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.1
- 95% interval
- 0.0 to 0.5
- Test series
- 12
- Test families
- 12 / 65
- Sources
- 2
- Share of the fixed test panel (informative)
- 18 %
Coding
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.2
- 95% interval
- 0.0 to 1.4
- Test series
- 3
- Test families
- 3 / 15
- Sources
- 1
- Share of the fixed test panel (informative)
- 11 %
Reasoning
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.1
- 95% interval
- 0.0 to 0.5
- Test series
- 1
- Test families
- 1 / 7
- Sources
- 1
- Share of the fixed test panel (informative)
- 7 %
Mathematics
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.0
- 95% interval
- 0.0 to 0.3
- Test series
- 1
- Test families
- 1 / 5
- Sources
- 1
- Share of the fixed test panel (informative)
- 20 %
Knowledge
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.1
- 95% interval
- 0.0 to 0.4
- Test series
- 1
- Test families
- 1 / 3
- Sources
- 1
- Share of the fixed test panel (informative)
- 33 %
Agents
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.1
- 95% interval
- 0.0 to 0.4
- Test series
- 1
- Test families
- 1 / 12
- Sources
- 1
- Share of the fixed test panel (informative)
- 8 %
General
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.4
- 95% interval
- 0.2 to 2.1
- Test series
- 1
- Test families
- 1 / 5
- Sources
- 1
- Share of the fixed test panel (informative)
- 20 %
Finance
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.3
- 95% interval
- 0.2 to 1.5
- Test series
- 2
- Test families
- 2 / 7
- Sources
- 1
- Share of the fixed test panel (informative)
- 29 %
Legal
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.2
- 95% interval
- 0.1 to 0.8
- Test series
- 1
- Test families
- 1 / 4
- Sources
- 1
- Share of the fixed test panel (informative)
- 25 %
Health
Index below 1 / 100
provisional
Published configurations differ within this comparison cohort. Treat the index as provisional.
- Standard error
- ± 0.1
- 95% interval
- 0.0 to 0.6
- Test series
- 1
- Test families
- 1 / 3
- Sources
- 1
- Share of the fixed test panel (informative)
- 33 %
Documented method · Included test families and measurements
Method version: benchmark-meta-score-v3
The index includes active measurements from reviewed primary sources when at least five models are compared in the same documented comparison slice across dataset, harness, agent, tools, and trial count. Each benchmark series and evaluator contributes one shared setup slice from the newest capture. When one source and series include several explicitly documented reasoning tiers, the highest documented tier (none through max) is selected, never the best result value. The explicit source label “Non-reasoning” is treated as none. Unknown or ambiguous tiers exclude the model from that series. Within each series, percentages and rates are logit-transformed, other units stay unchanged, and all values are then standardized. A one-factor model then estimates one shared strength per model and, per series, how sharply it separates the models. Area scores move away from that strength only as far as the model's own results in the area support. The overall score weights the core areas at 75% and finance, law and health at 25%. It requires at least eight test series from two sources and three areas. The 0 to 100 scale refers to a fixed reference group of 29 language models whose average sits at 50.
Published model-specific parameters can differ. This is not a controlled experiment; comparisons with such differences are marked provisional.
Unmeasured series are not entered as zero. Where a model has no own results in a core area, its overall strength stands in, and the note “estimated without own results in …” names the area. Fewer test series lead to a larger standard error.
The 95% interval follows from the standard error of 200 recalculations with randomly redrawn test families. It describes how precisely the recorded test series determine the value. It is no prediction for future tasks or for later recalculations with other participants.
The source count is not a count of independent experiments.
OECD handbook on composite indicatorsAll 12 test families and 12 of 12 raw measurements are listed here. The complete calculation for every model is available in the JSON export.
Test family: IFBench
General · Mean rank percentile of the family: 60.9 / 100Family weight in the fixed test panel (informative): 2.14 % · Coverage of this test family: 100 %
- Artificial Analysis
Published configurations differ within this cohort.
- Measurement:
- artificial-analysis-ifbench
- Series weight within the test family:
- 100 %
- Nominal contribution to the overall panel:
- 2.14 %
- Raw value:
- 68.0952380952381 %
- Metric:
- IFBench prompt-level loose accuracy (pass@1)
- Version:
- AA test set: 294 questions
- Participating models:
- 70
- Rank percentile within the cohort:
- 60.9 %
- Retrieved:
- 2026-10-19
- Comparison cohort:
- artificial-analysis-ifbench:AA test set: 294 questions:standalone:reasoning:tokens=2000:suffix=unspecified
- Documented comparison setup
- Publication cohort:
- ["artificial-analysis-ifbench","AA test set: 294 questions","aa-model-page-2026-10-19-8a022369738c"]
- Dataset type:
- artificial-analysis-evaluation
- Tool policy:
- Single-turn answer evaluated with the official IFBench source code
- Attempts:
- pass@1, 5 repeats per question
- Setup and model parameters
- Exact AA configuration: Kimi K2 Thinking (release kimi-k2-thinking, row kimi-k2-thinking, UUID bddebfd3-0a8d-47f5-b722-bc4c2ca5a5dc, reasoning=true, source reasoning-token configuration budget=2000, suffix=not specified). This configured value is not measured token consumption. Artificial Analysis runs 294 questions from the IFBench test set with five repeats per question. The official evaluator scores prompt-level accuracy in loose mode as pass@1.. Tool policy: Single-turn answer evaluated with the official IFBench source code. Attempts: pass@1, 5 repeats per question. Reasoning: reasoning enabled
Exact AA configuration: Kimi K2 Thinking (release kimi-k2-thinking, row kimi-k2-thinking, UUID bddebfd3-0a8d-47f5-b722-bc4c2ca5a5dc, reasoning=true, source reasoning-token configuration budget=2000, suffix=not specified). This configured value is not measured token consumption. Artificial Analysis runs 294 questions from the IFBench test set with five repeats per question. The official evaluator scores prompt-level accuracy in loose mode as pass@1.. Source model ID: artificial-analysis/bddebfd3-0a8d-47f5-b722-bc4c2ca5a5dc/kimi-k2-thinking/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Kimi. Model snapshot: Kimi K2 Thinking. Evaluator: Artificial Analysis. Dataset type: artificial-analysis-evaluation. Tool policy: Single-turn answer evaluated with the official IFBench source code. Attempts: pass@1, 5 repeats per question. Reasoning: reasoning enabled. Score display: 68.1%. Source ID: artificial-analysis. Evidence capture ID: aa-model-page-2026-10-19-8a022369738c. Source payload hash: 8a022369738cbe2a0e2f9bf486a2882865dbb0c6f1396dfec84059a802560e3c. Parser version: artificial-analysis-rsc-models-v2. Imported at: 2026-10-19
- Artificial Analysis
Test family: GPQA Diamond
Reasoning · Mean rank percentile of the family: 31.6 / 100Family weight in the fixed test panel (informative): 1.53 % · Coverage of this test family: 50 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-gpqa
- Series weight within the test family:
- 50 %
- Nominal contribution to the overall panel:
- 0.77 %
- Raw value:
- 78.536 %
- Metric:
- accuracy
- Version:
- 1
- Participating models:
- 77
- Rank percentile within the cohort:
- 31.6 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-gpqa:1:overall
- Documented comparison setup
- Publication cohort:
- vals-gpqa:1:overall
- Dataset type:
- public
- Setup and model parameters
- Temperature: 1. Maximum output tokens: 32768
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Temperature: 1. Maximum output tokens: 32768. Latency in seconds: 585.555. Standard error: 2.183. Cost per task in USD: 0.023072. Source ID: vals-ai. Evidence capture ID: vals-gpqa-2026-10-15-951eb5224097. Source payload hash: 951eb5224097733a9d09d06c14908ae2980e2e640e539f2c284877c5092d7077. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: SWE-bench Verified
Coding · Mean rank percentile of the family: 8.5 / 100Family weight in the fixed test panel (informative): 0.71 % · Coverage of this test family: 50 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-swebench
- Series weight within the test family:
- 50 %
- Nominal contribution to the overall panel:
- 0.36 %
- Raw value:
- 60.2 %
- Metric:
- accuracy
- Version:
- 1
- Participating models:
- 60
- Rank percentile within the cohort:
- 8.5 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-swebench:1:overall
- Documented comparison setup
- Publication cohort:
- vals-swebench:1:overall
- Dataset type:
- public
- Setup and model parameters
- Temperature: 0.6. Maximum output tokens: 32000
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Temperature: 0.6. Maximum output tokens: 32000. Latency in seconds: 1632.209. Standard error: 2.191. Cost per task in USD: 0.366026. Source ID: vals-ai. Evidence capture ID: vals-swebench-2026-10-15-7a4e2e848528. Source payload hash: 7a4e2e8485283caac500127419f68ec4bd5414c837ade114fc6afa5b2d34ccaa. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: Terminal-Bench
Coding · Mean rank percentile of the family: 35.7 / 100Family weight in the fixed test panel (informative): 0.71 % · Coverage of this test family: 17 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-terminal-bench-2
- Series weight within the test family:
- 16.67 %
- Nominal contribution to the overall panel:
- 0.12 %
- Raw value:
- 37.079 %
- Metric:
- accuracy
- Version:
- 2.0
- Participating models:
- 43
- Rank percentile within the cohort:
- 35.7 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-terminal-bench-2:2.0:overall
- Documented comparison setup
- Publication cohort:
- vals-terminal-bench-2:2.0:overall
- Dataset type:
- public
- Setup and model parameters
- Temperature: 1. Maximum output tokens: 128000
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-06-04. Temperature: 1. Maximum output tokens: 128000. Latency in seconds: 986.553. Standard error: 5.149. Cost per task in USD: 0.115876. Source ID: vals-ai. Evidence capture ID: vals-terminal-bench-2-2026-10-15-38e640d99525. Source payload hash: 38e640d995259778e531f048b2f2c7b3b281cb05b56be643ddc9dc53eb85adf7. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: CorpFin v2
Finance · Mean rank percentile of the family: 37.8 / 100Family weight in the fixed test panel (informative): 1.19 % · Coverage of this test family: 100 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-corp-fin-v2
- Series weight within the test family:
- 100 %
- Nominal contribution to the overall panel:
- 1.19 %
- Raw value:
- 60.567 %
- Metric:
- accuracy
- Version:
- 2
- Participating models:
- 75
- Rank percentile within the cohort:
- 37.8 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-corp-fin-v2:2:overall
- Documented comparison setup
- Publication cohort:
- vals-corp-fin-v2:2:overall
- Dataset type:
- private
- Setup and model parameters
- Temperature: 0.6. Maximum output tokens: 32000
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-08-12. Temperature: 0.6. Maximum output tokens: 32000. Latency in seconds: 93.297. Standard error: 0.962. Cost per task in USD: 0.038163. Source ID: vals-ai. Evidence capture ID: vals-corp_fin_v2-2026-10-15-7cf2217ab372. Source payload hash: 7cf2217ab3728d5a84bffce82aab9fc52d7f810b0cbe8c65b387df47f3d31c9b. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: LiveCodeBench v6
Coding · Mean rank percentile of the family: 15.6 / 100Family weight in the fixed test panel (informative): 0.71 % · Coverage of this test family: 100 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-lcb
- Series weight within the test family:
- 100 %
- Nominal contribution to the overall panel:
- 0.71 %
- Raw value:
- 63.145 %
- Metric:
- accuracy
- Version:
- 1
- Participating models:
- 78
- Rank percentile within the cohort:
- 15.6 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-lcb:1:overall
- Documented comparison setup
- Publication cohort:
- vals-lcb:1:overall
- Dataset type:
- public
- Setup and model parameters
- Temperature: 1. Maximum output tokens: 32768
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Temperature: 1. Maximum output tokens: 32768. Latency in seconds: 1150.624. Standard error: 1.157. Cost per task in USD: 0.033631. Source ID: vals-ai. Evidence capture ID: vals-lcb-2026-10-15-1a7c7dabc9d9. Source payload hash: 1a7c7dabc9d95a0182886b1edc42623987cac2095a0d407a5e334f37acdfa2f5. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: LegalBench
Legal · Mean rank percentile of the family: 33.3 / 100Family weight in the fixed test panel (informative): 2.08 % · Coverage of this test family: 100 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-legal-bench
- Series weight within the test family:
- 100 %
- Nominal contribution to the overall panel:
- 2.08 %
- Raw value:
- 80.201 %
- Metric:
- accuracy
- Version:
- 1
- Participating models:
- 79
- Rank percentile within the cohort:
- 33.3 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-legal-bench:1:overall
- Documented comparison setup
- Publication cohort:
- vals-legal-bench:1:overall
- Dataset type:
- public
- Setup and model parameters
- Temperature: 1. Maximum output tokens: 32768
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-10-13. Temperature: 1. Maximum output tokens: 32768. Latency in seconds: 43.991. Standard error: 0.459. Cost per task in USD: 0.002182. Source ID: vals-ai. Evidence capture ID: vals-legal_bench-2026-10-15-c8486ba756ba. Source payload hash: c8486ba756baa5cd3955d035edff67440e8db23e69d62ad1a9e18d15b8c4bc52. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: MedQA
Health · Mean rank percentile of the family: 61.5 / 100Family weight in the fixed test panel (informative): 2.78 % · Coverage of this test family: 100 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-medqa
- Series weight within the test family:
- 100 %
- Nominal contribution to the overall panel:
- 2.78 %
- Raw value:
- 92.592 %
- Metric:
- accuracy
- Version:
- 1
- Participating models:
- 40
- Rank percentile within the cohort:
- 61.5 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-medqa:1:overall
- Documented comparison setup
- Publication cohort:
- vals-medqa:1:overall
- Dataset type:
- public
- Setup and model parameters
- Temperature: 1. Maximum output tokens: 32768
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-04-16. Temperature: 1. Maximum output tokens: 32768. Latency in seconds: 225.305. Standard error: 0.239. Cost per task in USD: 0.006153. Source ID: vals-ai. Evidence capture ID: vals-medqa-2026-10-15-d67905978d93. Source payload hash: d67905978d932f64698ebe9e230d9519e5bae1c9d7176dda8379dc3874b6f90f. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: MGSM
Mathematics · Mean rank percentile of the family: 51.7 / 100Family weight in the fixed test panel (informative): 2.14 % · Coverage of this test family: 100 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-mgsm
- Series weight within the test family:
- 100 %
- Nominal contribution to the overall panel:
- 2.14 %
- Raw value:
- 90.146 %
- Metric:
- accuracy
- Version:
- 1
- Participating models:
- 30
- Rank percentile within the cohort:
- 51.7 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-mgsm:1:overall
- Documented comparison setup
- Publication cohort:
- vals-mgsm:1:overall
- Dataset type:
- public
- Setup and model parameters
- Temperature: 1. Maximum output tokens: 32000
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-01-09. Temperature: 1. Maximum output tokens: 32000. Latency in seconds: 48.41. Standard error: 1.83. Cost per task in USD: 0.000891. Source ID: vals-ai. Evidence capture ID: vals-mgsm-2026-10-15-1b87f953e52c. Source payload hash: 1b87f953e52c00bf4bdd90c3add26165f644eeb7a74a8382d2a0324ff0cc1040. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: MMLU-Pro
Knowledge · Mean rank percentile of the family: 28.6 / 100Family weight in the fixed test panel (informative): 3.57 % · Coverage of this test family: 100 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-mmlu-pro
- Series weight within the test family:
- 100 %
- Nominal contribution to the overall panel:
- 3.57 %
- Raw value:
- 81.068 %
- Metric:
- accuracy
- Version:
- 1
- Participating models:
- 78
- Rank percentile within the cohort:
- 28.6 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-mmlu-pro:1:overall
- Documented comparison setup
- Publication cohort:
- vals-mmlu-pro:1:overall
- Dataset type:
- public
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 94.771. Standard error: 0.396. Cost per task in USD: 0.002067. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: Agent Poker Bench
Agents · Mean rank percentile of the family: 12.5 / 100Family weight in the fixed test panel (informative): 0.89 % · Coverage of this test family: 100 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-poker-agent
- Series weight within the test family:
- 100 %
- Nominal contribution to the overall panel:
- 0.89 %
- Raw value:
- 1,011.6338984569229 rating
- Metric:
- TrueSkill rating
- Version:
- 1
- Participating models:
- 9
- Rank percentile within the cohort:
- 12.5 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-poker-agent:1:overall
- Documented comparison setup
- Publication cohort:
- vals-poker-agent:1:overall
- Dataset type:
- private
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2025-12-23. Latency in seconds: 71.72413331456877. Standard error: 0. Cost per task in USD: 0.02410503610648917. Source ID: vals-ai. Evidence capture ID: vals-poker_agent-2026-10-15-23821a0049c3. Source payload hash: 23821a0049c3ab8467573a74dac6c5037748e301e2abcd91885ec708168335af. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Test family: TaxEval v2
Finance · Mean rank percentile of the family: 38.2 / 100Family weight in the fixed test panel (informative): 1.19 % · Coverage of this test family: 100 %
- Vals AI
Published configurations differ within this cohort.
- Measurement:
- vals-tax-eval-v2
- Series weight within the test family:
- 100 %
- Nominal contribution to the overall panel:
- 1.19 %
- Raw value:
- 71.709 %
- Metric:
- accuracy
- Version:
- 2
- Participating models:
- 77
- Rank percentile within the cohort:
- 38.2 %
- Retrieved:
- 2026-10-15
- Comparison cohort:
- vals-tax-eval-v2:2:overall
- Documented comparison setup
- Publication cohort:
- vals-tax-eval-v2:2:overall
- Dataset type:
- private
- Setup and model parameters
- Temperature: 0.6. Maximum output tokens: 32000
Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-01. Temperature: 0.6. Maximum output tokens: 32000. Latency in seconds: 61.408. Standard error: 0.878. Cost per task in USD: 0.00913. Source ID: vals-ai. Evidence capture ID: vals-tax_eval_v2-2026-10-15-8b354a6ae848. Source payload hash: 8b354a6ae848f5bb25a2e8b061b5f576d5b9e0c11b941dc1e6bed078e6529cd6. Parser version: vals-astro-v3. Imported at: 2026-10-15
- Vals AI
Benchmark sources at a glance
Each evaluator measures different capabilities under its own conditions. Test series, composite indices and their components stay separate. The detailed results below retain their scales, configurations and source dates.
| Evaluator | Test series | Composite indices | Recorded active measurements |
|---|---|---|---|
| Artificial Analysis | 2 | 0 | 9 |
| Vals AI | 13 | 0 | 68 |
Coverage is not a quality ranking. Overlapping tests and multiple settings do not count as independent experiments. Vendor claims, historical and disputed results are excluded from this overview and remain labelled in the detailed evidence.
6 measurements from these sources still lack a verified test-family identity and do not increase the test-series count.
Page 1 of 9
Published benchmarks
Knowledge
MMLU-Pro
| Sub-task | Kimi K2 Thinking (This model) | GLM-5 | GLM-5.2 | GLM-5.1 | Gemini 3.1 Flash-Lite | DeepSeek-V3.2 |
|---|---|---|---|---|---|---|
| Biology | 90.24% | 94.28% (best value in this row) | 94% | 93.72% | 93.17% | 91.21% |
| Business | 86.57% | 90.75% | 91.89% (best value in this row) | 91.13% | 91.38% | 86.69% |
| Chemistry | 86.31% | 90.81% | 91.61% | 91.84% (best value in this row) | 90.99% | 87.19% |
| Computer Science | 83.9% | 87.56% | 90.24% (best value in this row) | 89.49% | 89.02% | 85.61% |
| Economics | 85.07% | 87.92% | 90.4% (best value in this row) | 88.98% | 89.34% | 86.73% |
| Engineering | 69.35% | 83.18% | 83.39% | 84.03% (best value in this row) | 83.7% | 77.61% |
| Health | 76.77% | 80.2% | 81.42% | 81.52% (best value in this row) | 80.32% | 78.85% |
| History | 72.18% | 77.43% | 76.9% | 78.42% (best value in this row) | 78.22% | 75.33% |
| Law | 63.31% | 72.84% | 72.03% | 74% | 75.02% (best value in this row) | 68.21% |
| Math | 91.64% | 94.82% | 95.86% (best value in this row) | 95.49% | 94.82% | 91.71% |
| Others | 80.41% | 84.2% | 83.66% | 84.52% (best value in this row) | 81.93% | 81.6% |
| Overall | 81.07% | 86.03% | 86.71% | 86.9% (best value in this row) | 86.24% | 83.06% |
| Philosophy | 79.56% | 83.17% | 84.17% (best value in this row) | 83.94% | 81.76% | 79.16% |
| Physics | 86.45% | 91.69% | 92.23% | 92.3% (best value in this row) | 91.99% | 88.3% |
| Psychology | 83.21% | 85.59% | 86.22% | 87.22% (best value in this row) | 85.71% | 84.59% |
higher is betterAs of October 15, 2026Source: Vals AIbars scaled within each row, not across rows
Methodology and context
- Metric
- accuracy
Recorded measurement details
MMLU-Pro: Biology (Show measurement, test conditions, and source)
- Source value
- 90.237
- Score
- 90.237
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- biology
- Task label
- Biology
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:biology
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 77.057. Standard error: 1.195. Cost per task in USD: 0.00166. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Business (Show measurement, test conditions, and source)
- Source value
- 86.565
- Score
- 86.565
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- business
- Task label
- Business
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:business
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 89.584. Standard error: 1.349. Cost per task in USD: 0.001802. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Chemistry (Show measurement, test conditions, and source)
- Source value
- 86.307
- Score
- 86.307
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- chemistry
- Task label
- Chemistry
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:chemistry
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 102.14. Standard error: 1.162. Cost per task in USD: 0.002417. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Computer Science (Show measurement, test conditions, and source)
- Source value
- 83.902
- Score
- 83.902
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- computer_science
- Task label
- Computer Science
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:computer_science
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 112.665. Standard error: 1.919. Cost per task in USD: 0.002135. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Economics (Show measurement, test conditions, and source)
- Source value
- 85.071
- Score
- 85.071
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- economics
- Task label
- Economics
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:economics
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 71.495. Standard error: 1.329. Cost per task in USD: 0.001543. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Engineering (Show measurement, test conditions, and source)
- Source value
- 69.35
- Score
- 69.35
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- engineering
- Task label
- Engineering
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:engineering
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 169.443. Standard error: 1.516. Cost per task in USD: 0.003996. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Health (Show measurement, test conditions, and source)
- Source value
- 76.773
- Score
- 76.773
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- health
- Task label
- Health
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:health
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 77.672. Standard error: 1.492. Cost per task in USD: 0.001618. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: History (Show measurement, test conditions, and source)
- Source value
- 72.178
- Score
- 72.178
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- history
- Task label
- History
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:history
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 87.832. Standard error: 2.302. Cost per task in USD: 0.002134. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Law (Show measurement, test conditions, and source)
- Source value
- 63.306
- Score
- 63.306
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- law
- Task label
- Law
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:law
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 145.75. Standard error: 1.466. Cost per task in USD: 0.003132. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Math (Show measurement, test conditions, and source)
- Source value
- 91.636
- Score
- 91.636
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- math
- Task label
- Math
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:math
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 76.794. Standard error: 0.92. Cost per task in USD: 0.002082. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Others (Show measurement, test conditions, and source)
- Source value
- 80.411
- Score
- 80.411
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- other
- Task label
- Others
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:other
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 67.872. Standard error: 1.33. Cost per task in USD: 0.001378. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Overall (Show measurement, test conditions, and source)
- Source value
- 81.068
- Score
- 81.068
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- overall
- Task label
- Overall
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:overall
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 94.771. Standard error: 0.396. Cost per task in USD: 0.002067. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Philosophy (Show measurement, test conditions, and source)
- Source value
- 79.559
- Score
- 79.559
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- philosophy
- Task label
- Philosophy
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:philosophy
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 85.402. Standard error: 1.812. Cost per task in USD: 0.001517. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Physics (Show measurement, test conditions, and source)
- Source value
- 86.451
- Score
- 86.451
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- physics
- Task label
- Physics
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:physics
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 98.509. Standard error: 1.014. Cost per task in USD: 0.0022. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
MMLU-Pro: Psychology (Show measurement, test conditions, and source)
- Source value
- 83.208
- Score
- 83.208
- Metric
- accuracy
- Unit
- %
- Benchmark version
- 1
- Category
- knowledge
- Task
- psychology
- Task label
- Psychology
- Direction
- Higher is better
- Comparison cohort
- vals-mmlu-pro:1:psychology
- Source type
- independent-evaluator
- Evaluator
- Vals AI
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Source model ID: kimi/kimi-k2-thinking. Source model name: Kimi K2 Thinking. Provider reported by source: Moonshot AI. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-01. Latency in seconds: 64.574. Standard error: 1.35. Cost per task in USD: 0.001324. Source ID: vals-ai. Evidence capture ID: vals-mmlu_pro-2026-10-15-dab025507f46. Source payload hash: dab025507f46c716dd8d145e90ec0a737d9c56219853f41e4b1cc3c5e8c222b4. Parser version: vals-astro-v3. Imported at: 2026-10-15
Page 1 of 9
Our tests
SVG short test
One SVG prompt and fixed output settings. Captures are shown only with 0 reported reasoning tokens or documented lack of reasoning controls. Identical test conditions do not make the models' internal compute identical.
Earlier tests
Pelican on a bicycle
A community classic for free-form SVG drawing.
Kimi K2 Thinking
API access was unavailable for this run. The model was not evaluated.
Reasoning (requested): high
Current attempt
- Run date
- Oct 5, 2026
- API provider
- openrouter
- Requested API model
- moonshotai/kimi-k2-thinking
- Reasoning setting
- high
- Pinned endpoint
- Google | moonshotai/kimi-k2-thinking-20251106
- Temperature
- Provider default
- Token limit including reasoning
- 8,192 tokens
- Test protocol
- community-visual-tests-v1
- Technical checks
- Not run
Not yet visually reviewed. A successful recording does not confirm correct physics or full compliance with the prompt.
Earlier attempts (2)
- Oct 2, 2026Output withheld for safetyReasoning setting: high
- Oct 1, 2026API request failedReasoning setting: high
Prompt and test conditions
Generate an SVG of a pelican riding a bicycle
Token limit including reasoning: 8,192 tokens
Task origin (Simon Willison)Popularity
20.6 / 100
Specifications and access
| Specification | Value and source |
|---|---|
| Weights | Open weightsSource |
| Access | API, LocalSource |
| Parameters | 1TSource |
| Active parameters | 32BSource |
| Context window | 262,144 tokensSource |
| License | modified MITSource |
| Parameters | 1,000 BSource |
| Active parameters | 32 BSource |
| Architecture | MoE with MLA and MoonViTSource |
| Input | Text, ImageSource |
| Output | TextSource |
| Reasoning mode | YesSource |
| Tool calling | YesSource |
| Availability | First-party Kimi API retired 2026-05-25 as part of the K2 series. (Released weights and third-party provider availability are separate from the first-party API lifecycle.)Source |
Sources and data date
| Type | Evidence and data date |
|---|---|
| Speed | Median across serving providers. Output speed after the first streaming chunk, not total response time. Classified as deprecated by Artificial Analysis. Artificial Analysis maintains only the default 10K-input workload there. Retrieved September 8, 2026. |
| Price | Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. |
| Input | Documented modalities. Retrieved September 8, 2026. |
| Output | Documented modalities. Retrieved September 8, 2026. |
| Research status | Research date September 8, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Unresolved | release-date Original wording No normalized release-date field in card. |
| Additional source | Model metadata |
| Additional source | Artificial Analysis (retrieved October 19, 2026) |
| Additional source | Artificial Analysis (retrieved October 19, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Vals AI (retrieved October 15, 2026) |
| Additional source | Model metadata (retrieved September 8, 2026) · Kimi API model list · Editorial description reviewed October 2, 2026 |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |