Prototype · closed beta
unOpen.AI
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
Kimi · Moonshot AI
Kimi K3.
Kimi K3 is a Moonshot AI model, released on 16 July 2026. It takes text, images and video and returns text, with a 1.05 million-token context window.
- Open weights
- 1.05M context
Specification
- Provider
- Moonshot AI
- Family
- Kimi
- Released
- 16 July 2026
- Catalogue id
- moonshotai/kimi-k3
- Context window
- 1.05 million tokens
- Maximum output
- 943,718 tokens
- Inputs and output
- Text, images and video in; text out
- Capabilities
- Tool calling, structured outputs and adjustable reasoning
- Weights
- Published: moonshotai/Kimi-K3
- Listed price
- $0.192 input, $15.00 output, $0.182 cached input per million tokens
Provider-published specifications as of 30 October 2026, not figures we measured. Model families move quickly — check the provider's own documentation before committing to a number.
Source: Kimi K3 in the OpenRouter catalogue, checked weekly. Confirm current terms in Moonshot AI's own documentation before you commit.
INDEPENDENT RATINGS
How Kimi K3 scores.
Measured by Artificial Analysis on its own evaluation suite, at every reasoning setting it tested.
Artificial Analysis Intelligence Index · Higher is better · Linear scale from 0 to 66.0
- Gemini 4 Carbon
- 59.4
- Claude Opus 5.5
- 58.4
- GPT-6 Astra
- 56.9
- Muse Wonder Full
- 49.7
- Kimi K3
- 17.4
The highest-rated model of each other family we cover, at its best reasoning setting.
Source: Artificial Analysis. Checked 30 October 2026. Measured independently by Artificial Analysis, not by QQuantum.ai or the providers.
BY REASONING SETTING
More thinking, more cost.
The same model tested at each reasoning effort. Higher settings usually score better and answer more slowly; measure which one your task needs.
| Criterion | Intelligence | Tokens / s | First token (s) | Blended $ / M |
|---|---|---|---|---|
| Low | 17.4 | — | — | $6.00 |
| Max | — | — | — | $6.00 |
BENCHMARKS
Benchmark by benchmark.
Score (%) at its best reasoning setting · Higher is better · Linear scale from 0 to 100.0
- GPQA Diamond (science)
- 84.2
- Humanity's Last Exam
- 25.0
- SciCode
- 52.7
- Terminal-Bench 4.0 (agentic coding)
- 12.6
- Terminal-Bench 2.1
- 82.4
- τ²-Bench Banking
- 41.7
- AA-LCR (long-context reasoning)
- 79.3
Each evaluation run by Artificial Analysis under its published methodology. Scores on public benchmarks can overstate performance on your own data.
Source: Artificial Analysis. Checked 30 October 2026. Measured independently by Artificial Analysis, not by QQuantum.ai or the provider.
What it costs
Price the whole answer.
Illustrative calculation at listed rates: a batch of 10 million input tokens and 1 million output tokens costs $16.92 on Kimi K3. A thousand requests of 2,000 input and 500 output tokens each costs $7.88. Calculation: input millions × input rate + output millions × output rate; it excludes reasoning tokens, retries, tools, tax and review time.
Compare cost per accepted result, not per token: replay the same work on two or three candidates and count the answers you would actually ship. Our model-routing approach sends each request to the cheapest model that passes.
THE REST OF THE FAMILY
Other Kimi models.
Every Kimi model in the catalogue. Our Kimi page covers where the family fits and how we evaluate it.
| Kimi K3 open | 16 Jul 2026 | — | — | 1.05M | $0.192 | $15.00 |
|---|---|---|---|---|---|---|
| Kimi K3.1 Pro | 23 Oct 2026 | 46.0 | 30 | 1.05M | $2.00 | $10.00 |
| Kimi K2.7 Code open | 12 Jun 2026 | — | 48 | 262k | $0.600 | $3.50 |
| Kimi K2.6 open | 20 Apr 2026 | — | — | 262k | $0.650 | $3.41 |
| Kimi K2.5 open | 27 Jan 2026 | — | — | 262k | $0.500 | $2.50 |
| Kimi K2 Thinking open | 6 Nov 2025 | — | — | 262k | $0.600 | $2.50 |
| Kimi K2 0905 open | 5 Sep 2025 | — | — | 262k | $0.600 | $2.50 |
| Kimi K2 0711 open | 11 Jul 2025 | — | — | 131k | $0.570 | $2.30 |
Intelligence and coding: Artificial Analysis indices at the model's best reasoning setting, and its measured output speed. Context and prices: the OpenRouter catalogue list price in USD per million tokens. Released: the provider's date where Artificial Analysis records one, otherwise the date the catalogue first listed it. Refreshed weekly; not measured by QQuantum.ai. Select a heading to sort.
Questions
Questions about Kimi K3.
When was Kimi K3 released?
Kimi K3 was released on 16 July 2026, according to Artificial Analysis.
What is the context window of Kimi K3?
1.05 million tokens, with up to 943,718 output tokens per response, as listed in the OpenRouter catalogue on 30 October 2026. A long window is not the same as reliable recall across it; evaluate retrieval on your own documents.
How much does Kimi K3 cost?
The catalogue lists $0.192 per million input tokens and $15.00 per million output tokens, and $0.182 per million cached input tokens (checked 30 October 2026). Moonshot AI's own rates, batch discounts and cloud marketplaces can differ.
Can Kimi K3 run on our own hardware?
Its weights are published (on Hugging Face as moonshotai/Kimi-K3), so it can be self-hosted subject to its licence. Check the licence and size the serving stack before buying hardware.
