Kimi K3 benchmarks, pricing and specs
Kimi K3 is a Moonshot Kimi model released on July 16, 2026 that takes text and image input and produces text. It costs $3 per 1M input tokens and $15 per 1M output tokens and accepts up to 1,048,576 tokens of context. It scores 1488 Elo on Arena Text, #7 of 39 current models (Arena, measured October 8, 2026).
- Per 1M input tokens
- $3
- Per 1M output tokens
- $15
- Context window
- 1.05M
- Arena Text
- 1488 Elo · #7/39
Data verified October 12, 2026 · Dataset 2026-10-12
Key facts
- Provider
- Moonshot Kimi
- Model family
- Kimi K3
- Release date
- July 16, 2026
- Status
- Current
- Access
- API and open weights
- API model id
- Not reported
- Context window
- 1,048,576 tokens
- Max output
- Not reported
- Input
- text, image
- Output
- text
- Task families
- text, multimodal
- Declared capabilities
- reasoning
Kimi K3 pricing
It is the #10 most expensive of 37 current text models priced per 1M input tokens.
Independent benchmark results
Scores come from independent evaluators that ran the model themselves, pinned to a dated snapshot. Rank is among current models in the same snapshot. Moonshot Kimi's own launch figures are listed under strengths with the publisher named, because they are not measured the same way.
| Benchmark | Score | Rank | Measured | Source |
|---|---|---|---|---|
| Arena Text | 1488 Elo | #7 of 39 | October 8, 2026 | Arena · 2026-10-12 |
| GPQA Diamond | 93.1% | #15 of 37 | October 12, 2026 | Epoch AI · 2026-10-12 |
| FrontierMath Tiers 1-3 (v2) | 72.2% | #19 of 37 | October 12, 2026 | Epoch AI · 2026-10-12 |
| SimpleQA Verified | 50.6% | #18 of 34 | October 12, 2026 | Epoch AI · 2026-10-12 |
Strengths, cautions and best uses
Strengths
- A 2.8-trillion-parameter mixture-of-experts model with about 104 billion active parameters, a 1M-token context window, and native vision
- Available both as a hosted API and as open weights: the checkpoint shipped 27 July 2026 under the Kimi K3 License as a native MXFP4 quantisation
- Introduces Kimi Delta Attention, a hybrid linear attention scheme, together with attention residuals
Cautions
- Hosted rates match Claude Sonnet 5 exactly at $3 / $15 while offering a larger context window, so the comparison turns on capability rather than price
- Running the open weights shifts cost to infrastructure, which the token rates recorded here do not represent
- Moonshot’s self-reported results place it ahead of Claude Opus 4.8 and GPT-5.5 but behind Claude Fable 5 and GPT-5.6 Sol; none of this was reproduced independently here
Best for
- Teams that want a frontier-scale model they can also self-host, with one architecture across both paths
- Long-context work at mid-tier hosted pricing
Alternatives to Kimi K3
The closest current textmodels by price. The cheapest model that scores as well or better is listed first; failing that, the cheapest one within 3% of the benchmark's scale.
Cheaper, same or higher score
Gemini 3.8 Flash
- 75% cheaper per 1M input tokens, $0.75 vs $3
- 9 points higher on Arena Text, 1497 vs 1488
- From Google
GPT-5.4
- 17% cheaper per 1M input tokens, $2.50 vs $3
- 13 points lower on Arena Text, 1475 vs 1488
- From OpenAI
Claude Opus 5.5
- 33% more per 1M input tokens, $4 vs $3
- 19 points higher on Arena Text, 1507 vs 1488
- From Anthropic
GPT-5.6 Sol
- 33% more per 1M input tokens, $4 vs $3
- 3 points lower on Arena Text, 1485 vs 1488
- From OpenAI
Compare Kimi K3 with
Articles about Kimi K3
Kimi K3 FAQ
How much does Kimi K3 cost?
Kimi K3 costs $3 per 1M input tokens and $15 per 1M output tokens at the standard tier. It is the #10 most expensive of 37 current text models priced per 1M input tokens. Source: OpenRouter, verified September 9, 2026.
What is the context window of Kimi K3?
Kimi K3 accepts up to 1,048,576 tokens of context. That ranks #10 of 44 current text models by context length.
How does Kimi K3 score on independent benchmarks?
Kimi K3 scores 1488 Elo on Arena Text, #7 of 39 current models (Arena, measured October 8, 2026), 93.1% on GPQA Diamond, #15 of 37 current models (Epoch AI, measured October 12, 2026) and 72.2% on FrontierMath Tiers 1-3 (v2), #19 of 37 current models (Epoch AI, measured October 12, 2026). These are independent evaluations, not Moonshot Kimi's own launch figures.
Is Kimi K3 open source or API only?
Kimi K3 is available both through Moonshot Kimi's API and as open weights you can run yourself.
What are the best alternatives to Kimi K3?
The closest current alternatives by price and measured performance are Gemini 3.8 Flash (75% cheaper per 1M input tokens, $0.75 vs $3), GPT-5.4 (17% cheaper per 1M input tokens, $2.50 vs $3), Claude Opus 5.5 (33% more per 1M input tokens, $4 vs $3) and GPT-5.6 Sol (33% more per 1M input tokens, $4 vs $3).
Sources
- Kimi platform model and pricing documentation · Moonshot AI · official · verified September 9, 2026
- OpenRouter models API catalogue · OpenRouter · independent · verified September 9, 2026
- Epoch AI Benchmarking Hub data (benchmark_data.zip) · Epoch AI · independent · verified October 12, 2026 · CC-BY-4.0
- Arena leaderboard dataset (lmarena-ai/leaderboard-dataset) · Arena · independent · verified October 12, 2026 · CC-BY-4.0
Cite this page
Van Data Team. "Kimi K3 Benchmarks, Pricing and Specs." AI Model Benchmark Dataset 2026-10-12. https://vandatateam.com/ai-model-benchmarks/kimi-k3
Reuse the data: JSON · CSV · or explore every model in the AI model benchmark comparison.
Production evaluation
Deciding whether Kimi K3 fits your workload?
Public benchmarks narrow the shortlist. We test the finalists on your own tasks, data and cost limits, then build the routing and review controls to run the winner in production.
- Evaluation cases and acceptance thresholds tied to real work
- Model routing, fallbacks and cost controls