Kimi K2.5

Lorphic's July 2026 explainer walks through the Kimi K2 model family, explicitly distinguishing K2, K2.5, K2.6, and K2.7 as architecturally related but with distinct capabilities and licensing. The piece confirms every K2-line model shares a 1-trillion-parameter MoE design with 32B active parameters, 384 routed experts The article situates Kimi K2.5 within the broader release timeline as an intermediate family member, useful for practitioners deciding which K2 variant to target. It frames the MoE architecture as the reason Kimi models can be priced competitively against dense models of comparable scale, since inference cost tracks ro

Moonshot AI released Kimi K2.5 in January 2026, an open-source model built on a Mixture-of-Experts architecture with 1 trillion total parameters and 32 billion activated per request. According to the Codecademy guide, it was trained on 15 trillion tokens that mixed visual and textual data from the start, letting vision The Codecademy article highlights Kimi K2.5's headline benchmark: 50.2% on Humanity's Last Exam at roughly 76% lower cost than Claude Opus 4.5, alongside claims of a 4.5x execution-time speedup from its parallel-agent approach. The Agent mode supports autonomous workflows with 200–300 tool calls. The piece positions K2

Constellation Research's January 27, 2026 coverage of the Kimi K2.5 launch documents four operating modes (K2.5 Instant, K2.5 Thinking, K2.5 Agent, and the K2.5 Agent Swarm beta) and explains that the Agent Swarm acts as an orchestrator that dynamically creates specialized sub-agents such as "AI Researcher," "Physics R On benchmarks, the article paraphrases Moonshot AI's release note claiming K2.5 delivers strong performance on HLE, BrowseComp, and SWE-Verified at a fraction of the cost of OpenAI's ChatGPT 5.2, though these are vendor-reported numbers whose harness, sampling, and inference budget are unspecified. The piece also situa

OpenRouter's catalog entry for moonshotai/kimi-k2.5 lists the model with a 262K context window, input/output modalities, and a release date of January 26, 2026, with headline pricing starting at $0.45 per 1M input tokens and $2.25 per 1M output tokens on DeepInfra (effective weighted averages around $0.1887 input / $2. A multi-provider routing table on the OpenRouter page spans DeepInfra, AtlasCloud, DigitalOcean, Moonshot AI, Amazon Bedrock, StreamLake, NovitaAI, Venice, Phala, and SiliconFlow with latency, throughput, uptime, and cache pricing; notably, Meganova does not appear in this provider list, so the data cannot directly cor

Per the official Ollama library page, Kimi K2.5 is an open-source, native multimodal agentic model from Moonshot AI built through continual pretraining on approximately 15 trillion mixed visual and text tokens atop Kimi-K2-Base, integrating vision and language understanding with instant and thinking modes. The page hig Architecturally, the Ollama description frames K2.5 around cross-modal reasoning, visual knowledge, and agentic tool use grounded in visual inputs, which aligns with Meganova's OpenAI-compatible inference endpoint and gives developers a concrete picture of what the model can be invoked for (image-grounded tool use, vis