Kimi K2 Models by Moonshot AI
Last refreshed 2026-10-19. Next refresh: weekly.
Details
Capabilities
Links
WebsiteAbout
Kimi K2 is Moonshot AI's frontier mixture-of-experts family for agentic coding and long-context reasoning. The line spans 8 models, from the original Kimi K2 release through Kimi K2 Thinking and the current flagship Kimi K2.6, with 1T total parameters and 32B active per forward pass. Kimi K2.6 scores 1454 on Chatbot Arena, 92 on HumanEval, 89.6 on LiveCodeBench, and 80.2 on SWE-bench Verified, putting it near Claude Opus 4.7 and DeepSeek V4 Pro for coding tasks at lower listed API prices.
Decision facts
Compare Against Top Competitors
Frontier coding comparison set for teams deciding between Kimi K2.6, Claude, DeepSeek V4, and GPT-5-class models. Scores come from existing benchmark seed data; "-" means this site has no local score for that benchmark yet.
| Model | Context | Input / 1M | Chatbot Arena | SWE-bench Verified | LiveCodeBench | HumanEval |
|---|---|---|---|---|---|---|
| Kimi K2.6family pick | 256K | $0.800/1M | 1,462 | 80.2 | 89.6 | 92 |
| Claude Opus 4.7 | 1M | - | 1,503 | 87.6 | - | - |
| DeepSeek V4 Pro | 1M | - | 1,456 | 80.6 | 93.5 | 76.8 |
| GPT-5.5 | 1.05M | - | 1,488 | 82.6 | - | 94.2 |
| Claude Sonnet 4.6 | 1M | - | 1,459 | 79.6 | 80 | 98 |
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs 256K context, 1K parameters, and JSON / Tool use.
Use when the workload needs 128K context, reasoning, and structured outputs.
Use when the workload needs reasoning, 256K context, and structured outputs.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Kimi K2 0905 Preview | Use when the workload needs 256K context, 1K parameters, and JSON / Tool use. | 2025-09 | 256K context1K parametersJSON / Tool use | Current |
| Kimi K2 Instruct | Use when the workload needs 128K context, reasoning, and structured outputs. | 2025-09 | 128K contextreasoningstructured outputs | Current |
| Kimi K2 Instruct 0905 | Use when the workload needs 128K context. | 2025-09 | 128K context | Current |
| Kimi K2 Thinking | Use when the workload needs reasoning, 256K context, and structured outputs. | 2025-01 | reasoning256K contextstructured outputs | Current |
Release Timeline
7 release groupsSpecifications(11 models)
| Model | Released | Context | Parameters | Vision | Multimodal | Reasoning | JSON / Tool use | Structured Outputs |
|---|---|---|---|---|---|---|---|---|
| Kimi K2 0905 Preview | 2025-09 | 256K | 1K | No | No | No | Yes | No |
| Kimi K2 Instruct | 2025-09 | 128K | 1T total, 32B active (MoE) | No | No | Yes | No | Yes |
| Kimi K2 Instruct 0905 | 2025-09 | 128K | 1T total, 32B active (MoE) | No | No | No | No | No |
| Kimi K2 Thinking | 2025-01 | 256K | 1T (32B active) | No | No | Yes | No | Yes |
Available From(12 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| Kimi K2 Instruct | Vercel AI Gateway | $0.57 | $2.3 | Serverless |
| Kimi K2 Instruct | Novita AI | $0.57 | $2.3 | Serverless |
| Kimi K2 Instruct | Fireworks AI | $0.6 | $2.5 | Serverless |
| Kimi K2 Instruct 0905 | Fireworks AI | $0.6 | $2.5 | Serverless |
| Kimi K2 Thinking | Fireworks AI | $0.6 | $2.5 | Serverless |
| Kimi K2 Thinking | GCP Vertex AI | $0.6 | $2.5 | Serverless |
| Kimi K2 Thinking | AWS Bedrock | $0.6 | $2.5 | Serverless |
| Kimi K2 Thinking | OpenRouter | $0.6 | $2.5 | Serverless |
| Kimi K2 Thinking | Vercel AI Gateway | $0.6 | $2.5 | Serverless |
| Kimi K2 Thinking | Novita AI | $0.6 | $2.5 | Serverless |
| Kimi K2 0905 Preview | Novita AI | $0.6 | $2.5 | Serverless |
| Kimi K2 Instruct | Featherless | $0.6 | $2.5 | Serverless |
| Kimi K2 Instruct 0905 | OpenRouter | $0.6 | $2.5 | Serverless |
| Kimi K2 Instruct | Together AI | $1.2 | $4.5 | Serverless |
Popular comparisons in this family
Comparisons
- GLM-5.2 vs Kimi K2.6
- Claude Sonnet 5 vs Kimi K2.6
- DeepSeek V4 Flash vs Kimi K2.6
- Claude Opus 4.8 vs Kimi K2.6
- Claude Opus 4.7 vs Kimi K2 Thinking
- Kimi K2.6 vs Kimi K2.5
- Kimi K2.6 vs Claude Opus 4.7
- Kimi K2.6 vs DeepSeek V4 Pro
Models(11)
Kimi K2 0905 Preview
Kimi K2 Instruct
Kimi K2 Instruct 0905
Kimi K2 Thinking