Intelligence
33 scoreKimi · Flagships Analysis
Kimi K2 Thinking
Intelligence, Performance & Price Analysis
Canonical slug: kimi-k2-thinking · Canonical model registry at build time
Speed
160.2 output tokens/secLatency
1.26s TTFTInput Price
$0.60 / 1M tokensOutput Price
$2.50 / 1M tokensExecutive Assessment
Routing Verdict & Tradeoffs
Ethen Routing Verdict
Frontier research / complex reasoning
Kimi K2 Thinking scores 33 (estimated) on the Artificial Analysis Intelligence Index, placing it above average among other open weight models of similar size (median: 25). Kimi K2 Thinking generates output at 160.2 tokens per second (based on the median across providers serving the model), which is well above average compared to other open weight models of similar size (median: 61.7 t/s). Kimi K2 Thinking costs $0.60 per 1M input tokens (somewhat higher than average, median: $0.59) and $2.50 per 1M output tokens (somewhat higher than average, median: $2.20), based on the median across providers serving the model.
Reserve this model for the hardest requests; simpler or repetitive work should stay on cheaper routes.
Task Fit Assessment
| Workload | Rating | Notes |
|---|---|---|
| Complex reasoning & agentic workflows | optimal | Intelligence score 33 places this model among top performers. Suitable for multi-step analysis and agentic loops. |
| High-volume chat & customer-facing | optimal | Output speed 160.2 tokens/sec and capable intelligence make this suitable for real-time chat at scale. |
| Latency-sensitive applications | optimal | TTFT 1.26s — adequate latency for most interactive use cases. |
| Cost-sensitive pipelines | viable | Output pricing at $2.50 is premium. Route high-volume simple tasks to cheaper alternatives. |
Cost Pressure Analysis
Medium
Pricing is moderate — input $0.60, output $2.50. Costs accumulate at volume but are manageable for valuable tasks.
Technical Specifications
Architecture and Limits
| Specification | Value | Validation Authority |
|---|---|---|
| Model type | Open weights | inferred |
| Reasoning | Yes | faq |
| Input modalities | Kimi K2 Thinking supports text input. | faq |
| Output modalities | Kimi K2 Thinking supports text output. | faq |
| Context window | 260k tokens | faq |
| Open weights / source | Yes, Kimi K2 Thinking is open weights. The model weights are publicly available and can be downloaded for self-hosting. | faq |
| Parameters | Kimi K2 Thinking has 1 trillion parameters (32 billion active). | faq |
| Active parameters | Kimi K2 Thinking is a Mixture of Experts (MoE) model with 1 trillion total parameters, but only 32 billion active parameters are used during inference. | faq |
| License | Kimi K2 Thinking is released under the Modified MIT License license. This license allows commercial use. | faq |
| API availability | Yes, Kimi K2 Thinking is available via API through 6 providers. | faq |
Evidence charts
Profile visualizations
Charts are the same canonical evidence cards previously published for this slug, contained inside the D18D page grammar.
AA-Omniscience Index
AA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct. · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Bars intentionally show many models at once so the page feels like a benchmark report rather than a dashboard.
Artificial Analysis Intelligence Index by Open Weights / Proprietary
Artificial Analysis Intelligence Index v4.1 incorporates 9 evaluations: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Artificial Analysis Intelligence Index
Artificial Analysis Intelligence Index v4.1 incorporates 9 evaluations: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Artificial Analysis Openness Index: Score
Openness Index assesses model openness on a 0 to 100 normalized scale (higher is more open) · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Bars intentionally show many models at once so the page feels like a benchmark report rather than a dashboard.
Intelligence
Artificial Analysis Intelligence Index · Higher is better · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Output Speed
Output tokens per second · Higher is better · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Speed should be read together with latency and cost; fast output does not always mean better user-perceived performance.
Speed
Output tokens per second · Higher is better · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Speed should be read together with latency and cost; fast output does not always mean better user-perceived performance.
End-to-End Response Time
Seconds to output 500 tokens, including reasoning model 'thinking' time · Lower is better · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Bars intentionally show many models at once so the page feels like a benchmark report rather than a dashboard.
Latency: Time To First Answer Token
Seconds to first answer token received · Accounts for reasoning model 'thinking' time · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Latency includes thinking / reasoning time when applicable. Lower is better for interactive use cases.
Context Window
Context window: tokens limit · Higher is better · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Larger context windows enable complex multi-turn tasks but may increase latency and cost for long inputs.
Cost per Intelligence Index Task
Weighted average cost (USD) per Artificial Analysis Intelligence Index task, segmented by token type. Lower is better · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Ethen Gateway should route away from this model when the task is simple, repetitive, or likely to generate many output tokens.
Cost per Task
Weighted average cost (USD) per Intelligence Index task · Lower is better · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Ethen Gateway should route away from this model when the task is simple, repetitive, or likely to generate many output tokens.
Cost to Run Artificial Analysis Intelligence Index
Cost (USD) to run all evaluations in the Artificial Analysis Intelligence Index · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Ethen Gateway should route away from this model when the task is simple, repetitive, or likely to generate many output tokens.
Pricing: Cache Hit, Input, and Output
Price (USD per M Tokens) · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Bars intentionally show many models at once so the page feels like a benchmark report rather than a dashboard.
Time per Intelligence Index Task
Weighted average decode time (minutes) per task; excludes TTFT and overhead time · Lower is better · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Model Size: Total and Active Parameters
Comparison between total model parameters and parameters active during inference · Evaluation results measured independently by Artificial Analysis
Retrieved 2026-07-08
Chart source and provenance are listed in Methodology & sources below.
- Source:
- Published benchmark dataset
- Retrieval date:
- 2026-07-08
- Methodology:
- Independent test run by Artificial Analysis on dedicated hardware.
Larger models generally perform better but cost more to serve. Active parameter count matters for inference efficiency.
Methodology
Methodology & Provenance
This page is rendered from the normalized profile and page JSON for Kimi K2 Thinking.
Benchmark values are preserved as normalized; only layout, disclosure ordering, and typography are adjusted for readability.
Frequently Asked Questions
Model FAQs & Technical Disclosures
Kimi K2 Thinking was released on November 6, 2025.
Kimi K2 Thinking was created by Kimi.
Kimi K2 Thinking scores 33 (estimated) on the Artificial Analysis Intelligence Index, placing it above average among other open weight models of similar size (median: 25).
Kimi K2 Thinking generates output at 160.2 tokens per second (based on the median across providers serving the model), which is well above average compared to other open weight models of similar size (median: 61.7 t/s).
Kimi K2 Thinking has a time to first token (TTFT) of 1.26s (based on the median across providers serving the model), which is very competitive compared to other open weight models of similar size (median: 2.01s).
Kimi K2 Thinking costs $0.60 per 1M input tokens (somewhat higher than average, median: $0.59) and $2.50 per 1M output tokens (somewhat higher than average, median: $2.20), based on the median across providers serving the model.
Kimi K2 Thinking costs $0.60 per 1M input tokens and $2.50 per 1M output tokens (based on the median across providers serving the model). For a blended rate (7:2:1 cache hit/input/output ratio), this is $0.79 per 1M tokens. Pricing may vary by provider.
Yes, Kimi K2 Thinking is a reasoning model. It uses extended thinking or chain-of-thought reasoning to work through complex problems before providing an answer.
Kimi K2 Thinking supports text input.
Kimi K2 Thinking supports text output.
No, Kimi K2 Thinking does not support image input. It can only process text.
No, Kimi K2 Thinking is not multimodal. It only supports text input.
Kimi K2 Thinking has a context window of 260k tokens. This determines how much text and conversation history the model can process in a single request.
Yes, Kimi K2 Thinking is open weights. The model weights are publicly available and can be downloaded for self-hosting.
Kimi K2 Thinking has 1 trillion parameters (32 billion active).
Kimi K2 Thinking is a Mixture of Experts (MoE) model with 1 trillion total parameters, but only 32 billion active parameters are used during inference.
Kimi K2 Thinking is released under the Modified MIT License license. This license allows commercial use.
Kimi K2 Thinking achieves a score of 33 on the Artificial Analysis Intelligence Index. This composite benchmark evaluates models across reasoning, knowledge, mathematics, and coding.
Yes, Kimi K2 Thinking is available via API through 6 providers.
Kimi K2 Thinking is available through 6 API providers.