Moonshot AI

The Moonshot AI Kimi Google Play listing shows a cumulative 5 million-plus download threshold and a September 23, 2026 update. This observable consumer distribution measure is distinct from recorded model specifications; downloads do not establish monthly active users, paying customers or production enterprise deployment.

Moonshot AI released Kimi K3, a 2.8T-parameter open model with native vision, a 1M-token context window, agentic coding, long-horizon task execution, and an agent swarm architecture.

Moonshot AI released Kimi K3, a 2.8T-parameter open model with native vision, a 1M-token context window, agentic coding, long-horizon task execution, and an agent swarm architecture.

Moonshot's Kimi K2 technical release documents MuonClip, including attention-logit control for stable large-scale Muon optimization. The training-method contribution is independent of existing K2.5 multimodal swarm-orchestration evidence.

Kimi K2.5 introduced native multimodal capabilities and a self-directed agent swarm capable of large-scale tool orchestration, supporting frontier-level agent and model research.

Kimi K2.5 introduced native multimodal capabilities and a self-directed agent swarm capable of large-scale tool orchestration, supporting frontier-level agent and model research.

Kimi Business documents organizational seat allocation by administrators, business contracts and invoicing support, and a policy excluding business data from model training. This provides distinct evidence of enterprise procurement and operating readiness.

Kimi models are available through Kimi.com, apps, API and coding products, with K3 weights also released, showing a broad product and developer distribution footprint beyond a research-only lab.

Kimi models are available through Kimi.com, apps, API and coding products, with K3 weights also released, showing a broad product and developer distribution footprint beyond a research-only lab.

Moonshot AI open-sources Kimi Vendor Verifier tests for decoding constraints, visual preprocessing, long-output inference and tool-call JSON Schema accuracy, and describes collaboration with vLLM, SGLang and KTransformers communities. Concrete cross-provider verification tooling supports inferred portable inference-engineering skills.

Kimi K3 and K2.5 emphasize long-horizon coding, multimodal reasoning, tool use, model serving and large-scale agent orchestration, skills with broad portability across frontier AI engineering roles.

Kimi K3 and K2.5 emphasize long-horizon coding, multimodal reasoning, tool use, model serving and large-scale agent orchestration, skills with broad portability across frontier AI engineering roles.