Vendor · Language models
Moonshot AI has 12 models in the directory. The strongest is Kimi K3.1 Pro: Rank 10 to 13 of 95 in the overall index.
In the directory: 12 language models. 10 of them with open weights that can run locally.
Every model there ever was. Models with a profile lead to every benchmark, price and source.
| Kimi K3.1 Pro | Oct 23, 2026 | Active | 80Rank 10 to 13 of 95 | $2 / $10 | 36 | 1.05M | below 1 |
| Kimi K2.8 Preview | Sep 11, 2026 | Preview | n/a | n/a | n/a | 1.05M | below 1 |
| Kimi K3 | Jul 16, 2026 | Active | 54Rank 21 to 29 of 95 | $3 / $15 | 55 | 1.05M | 68 |
| Kimi K2.7 Code Highspeed | Jun 12, 2026 | Active | n/a | n/a | n/a | 262K | 25 |
| Kimi K2.7 Code | June 2026 | Active | 10Rank 49 to 59 of 95 | $0.95 / $4 | 66 | 262K | 25 |
| Kimi K2.6 | Apr 20, 2026 | Active | 15Rank 45 to 51 of 95 | $0.95 / $4 | 58 | 262K | 47 |
| Kimi K2.5 | January 2026 | Open source | 2Rank 59 to 69 of 95 | $0.57 / $2.85 | 88 | 262K | 47 |
| Kimi K2 Thinking Turbo | Nov 6, 2025 | Active | n/a | n/a | n/a | 262K | 7 |
| Kimi K2 Thinking | Nov 4, 2025 | Open source | below 1Rank 74 to 80 of 95 | $0.60 / $2.50 | 121 | 262K | 21 |
| Kimi K2 0905 | Sep 5, 2025 | Open source | n/a | $0.60 / $2.50 | 47 | 262K | 31 |
| Kimi K2 0711 | Jul 11, 2025 | Open source | n/a | $0.57 / $2.30 | 40 | 131K | 35 |
| Kimi K2 | July 2025 | Active | n/a | n/a | 46 | 131K | 37 |
Language model · Oct 23, 2026
Kimi K3.1 Pro is Moonshot's reasoning model. Its context window reaches 1,048,576 tokens on the Pro or Allegretto Kimi Code tier; lower tiers have a smaller limit. Kimi Code requires Thinking and offers Low, High, and Max effort. Moonshot says the Highspeed variant has the same model quality, so it is not a separate model profile.
Language model · Jul 16, 2026
What is distinctive about Kimi K3 is its native visual architecture with KDA and AttnRes. Moonshot describes 2.8 trillion total parameters, 104 billion active parameters, 896 experts with 16 selected per step, and a 1,048,576-token context window. Text, images, and video belong to the same model.
Language model · Jun 12, 2026
Kimi K2.7 Code Highspeed is described in Kimi’s official catalog as the fast serving variant of Kimi K2.7 Code. The catalog gives the coding family a 256,000-token context window, about 180 tokens per second, and up to 260 tokens per second in short contexts. Publisher sources identify no separate Highspeed model card or weight release.
Language model · June 2026
What is distinctive about Kimi K2.7 Code is its explicit focus on long software projects. Moonshot combines 1 trillion total parameters and 32 billion active parameters with a 256,000-token context window, tool calls, and multimodal input for text, images, and experimental video.
Language model · Apr 20, 2026
New in Kimi K2.6 are MoonViT for image and video input and an emphasis on long-horizon coding and agent swarms. Moonshot describes 1 trillion total parameters, 32 billion active parameters, and a 262,144-token context window. Thinking and non-thinking modes can be used separately.
Language model · January 2026
What is distinctive about Kimi K2.5 is the combination of the K2 mixture-of-experts design with MoonViT. The open multimodal model accepts text, images, and experimental video with 1 trillion total parameters, 32 billion active parameters, and a 262,144-token context window.
Language model · Nov 6, 2025
Kimi K2 Thinking Turbo is listed as a discontinued API model in Kimi’s official catalog. Publisher sources identify no separate Turbo model card or weight release; they point to the distinct Thinking checkpoint with 1 trillion total parameters, 32 billion active parameters, long context, and tool calls.
Language model · Nov 4, 2025
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.
Language model · Sep 5, 2025
New in Kimi K2 0905 is the 262,144-token context window compared with Kimi K2 0711. The underlying K2 mixture-of-experts design remains an open model with 1 trillion total parameters and 32 billion active parameters for dialogue, coding, and tool use.
Language model · Jul 11, 2025
Kimi K2 0711 is the dated Instruct release from July 11, 2025. Moonshot’s K2 mixture-of-experts design has 1 trillion total parameters and 32 billion active parameters; this checkpoint uses a 131,072-token context window.
Language model · July 2025
Kimi K2 is Moonshot’s open K2 Instruct release. The repository describes 1 trillion total parameters, 32 billion active parameters, and about 131,000 tokens of context. This entry represents the foundational K2 Instruct release.
Derived from the overall index, the API price and the measured speed of the current models.
The strongest choice
Kimi K3.1 Pro
The vendor's highest overall index
Value for money
Kimi K2.6
Costs 43% of Kimi K3.1 Pro, overall index 15 instead of 80
The fastest
Kimi K2 Thinking
121 tokens/s, a typical model manages 98
The 20 best language models, with the models from Moonshot AI highlighted. “All” and the filter show every model in the index.
20 of 95 models
Provider
Weights
Price (blended per 1M tokens)
Released
Anthropic
OpenAI
Meta
Moonshot AI
DeepSeek
xAI
Z.ai
Xiaomi
Alibaba
StepFun
Mistral AI
MiniMax
Thinking Machines Lab
Inception
| Vendor | Models | Model types | Best language model | Top 10 | Open weights | New in 2026 |
|---|---|---|---|---|---|---|
| Anthropic | 33 | 33 language models | Claude Opus 5.5 (Rank 1 to 3) | 4 | 0 | 13 |
| OpenAI | 63 | 46 language models, 4 image models, 2 video models, 9 audio models, 2 specialist models | GPT-6 Astra (Rank 4 to 8) | 2 | 4 | 27 |
| 66 | 43 language models, 4 image models, 6 video models, 11 audio models, 2 specialist models | Gemini 4 Carbon (Rank 1 to 5) | 2 | 19 | 26 | |
| Meta | 32 | 29 language models, 1 image model, 1 video model, 1 specialist model | Muse Wonder Full (Rank 6 to 9) | 2 | 26 | 7 |
| DeepSeek | 23 | 23 language models | DeepSeek V4.1 Flash 1017 (Rank 13 to 20) | 0 | 23 | 9 |
| Moonshot AI | 12 | 12 language models | Kimi K3.1 Pro (Rank 10 to 13) | 0 | 10 | 7 |
| xAI | 16 | 12 language models, 2 image models, 2 video models | Grok 4.7 (Rank 18 to 24) | 0 | 1 | 12 |
By release date. Grey: retired models.
7 releases since January 2026, by release date.
Prices per 1M tokens for the current language models, according to Moonshot AI or the provider linked in each model profile.
| Current model | Input | Output | Context |
|---|---|---|---|
| Kimi K3.1 Pro | $2 | $10 | 1.05M |
| Kimi K3 | $3 | $15 | 1.05M |
| Kimi K2.6 | $0.95 | $4 | 262K |
| Kimi K2.7 Code | $0.95 | $4 | 262K |
| Kimi K2.5 | $0.57 | $2.85 | 262K |
| Kimi K2 Thinking | $0.60 | $2.50 | 262K |
| Kimi K2 0711 | $0.57 | $2.30 | 131K |
| Kimi K2 0905 | $0.60 | $2.50 | 262K |
Monthly prices in Germany incl. VAT; “≈” is converted from US dollars. The vendor's models are often also part of other providers' plans, see the AI subscription comparison.
| Plan | Per month | Note |
|---|---|---|
| Kimi Free | free | |
| Kimi Plus | ≈ €20.24 | with credits |
| Kimi Pro | ≈ €41.54 | |
| Kimi Max | ≈ €105.45 | |
| Kimi Ultra | ≈ €211.97 |
The directory lists 12 Kimi models from Moonshot AI: 12 language models. It counts every model Moonshot AI ever released, including retired ones. 11 are currently available, the newest being Kimi K3.1 Pro, Kimi K3 and Kimi K2.7 Code Highspeed.
Kimi K3.1 Pro. It is at rank 10 to 13 of 95 in the overall index, which combines independent benchmarks from several sources.
Kimi K3.1 Pro, released on Oct 23, 2026. The timeline below shows every release by date.
Through the API per 1M tokens (input / output): Kimi K2.5 $0.57 / $2.85 and Kimi K3.1 Pro $2 / $10. The prices of all 8 current language models are in the price table. As a subscription, Kimi starts at €20.24 per month incl. VAT (Kimi Plus).
Yes, with the Kimi Free plan. The cheapest paid plan is Kimi Plus at €20.24 per month incl. VAT.
In the overall index, no. The best OpenAI model, GPT-6 Astra (rank 4 to 8 of 95), is ahead of Kimi K3.1 Pro (rank 10 to 13 of 95). Single tasks can differ.
Yes, 10 of the 12 models have open weights. Each model profile lists how much memory it needs.
Kimi K2 Thinking with 121 tokens per second according to Artificial Analysis. A typical model manages 98.
Up to 1.05 M tokens (≈ 1,570 A4 pages), which 3 of the current models reach, including Kimi K3.1 Pro. Every model's value is in the model list.
Data as of Nov 1, 2026