Kimi K2.5
#34 in Frontier LLMsmoonshot-ai · seit 2026-01-27 · 46× · zuletzt 02. Juli 2026
35
Momentum
Kimi K2.5 is an open-source native multimodal agentic model by Moonshot AI, built via continual pretraining on approximately 15 trillion mixed visual and text tokens on top of Kimi-K2-Base. It is a Mixture-of-Experts (MoE) model with 1 trillion total parameters and 32 billion active parameters per forward pass. The model natively supports image and video input, Thinking and Instant modes, and an agentic Agent Swarm capability coordinating up to 100 sub-agents. It is accessible via the official Moonshot platform, Hugging Face (open weights), Amazon Bedrock, and OpenRouter.
Momentum trend
04.04.03.07.
Features
| Key Benchmark (%) | HLE-Full with tools: 50.2% (text: 51.8%, image: 39.8%); SWE-Bench Verified: 76.8%; AIME 2025: 96.1%; GPQA-Diamond: 87.6%; BrowseComp (Agent Swarm): 78.4%; LiveCodeBench v6: 85% |
| Context Window (Tokens) | 256,000 tokens (official per Moonshot API docs and Amazon Bedrock); some third parties configure up to 262,144 tokens |
| License | Modified MIT License (open weights; commercial use requires attribution above 100 million monthly active users or $20 million monthly revenue) |
| Platform | kimi.com (web & app), Moonshot API (platform.moonshot.ai / api.moonshot.ai), Hugging Face (self-hosting), Amazon Bedrock, OpenRouter; inference engines: vLLM, SGLang, KTransformers |
| Price | Free access via kimi.com (with usage limits); API access paid (pay-per-token); open weights free via Hugging Face under Modified MIT License |
| Price per 1M Tokens | $0.60 / 1M input tokens, $3.00 / 1M output tokens (Moonshot API, cache miss); cache hit price: $0.10 / 1M input tokens |
| Release Date | January 27, 2026 |
