Kimi

What is Kimi?

Kimi is the consumer AI assistant from Moonshot AI, one of China's leading AI labs. You can use it on the web at kimi.com, through mobile apps on iOS and Android, via the Moonshot / Kimi API for developers, and through the Kimi Code CLI for terminal-based coding workflows. Kimi shipped as a chatbot with bilingual Chinese–English support and quickly became known for long-context document handling and aggressive pricing.

Under the hood, Kimi is powered by the Kimi K2.5 model family, released in January 2026. K2.5 is a Mixture-of-Experts (MoE) architecture with roughly 1 trillion total parameters but only about 32 billion active per token, which keeps inference cost low while preserving strong reasoning and coding performance. It ships with a 256K-token context window — long enough to load an entire book, a large codebase, or a stack of research papers into one conversation — and Moonshot also publishes Kimi K2 Thinking, a reasoning-tuned variant for harder problems, plus an "Agent Swarm" mode that coordinates multiple specialized sub-agents to tackle complex tasks in parallel.

What sets Kimi apart in 2026 isn't just long context — it's the price. API pricing for Kimi K2.5 is roughly $0.60 per million input tokens and $2.50 per million output tokens, which dramatically undercuts frontier Western models from OpenAI and Anthropic. Combined with automatic context caching (which can reduce input costs by up to 75% on repeated prompts), Kimi has become one of the cheapest ways to access a capable long-context model. The consumer Kimi Chat at kimi.com remains free, with paid tiers unlocking higher usage for features like Deep Research and the OK Computer agent.

Kimi Pricing

Kimi Chat (web and mobile) — Free: The Adagio free plan at kimi.com offers unlimited basic conversations with Kimi K2.5, file uploads, and the 256K context window. Intensive features like Deep Research queries and OK Computer agent tasks are rate-limited on the free tier.

Paid Kimi Chat plans: Moonshot offers higher-tier consumer plans that unlock more Deep Research runs, more agent executions, and larger file limits. Exact tier pricing varies by region and launch market — see the official Kimi pricing page for current monthly rates.

Kimi K2.5 API: Roughly $0.60 per million input tokens and $2.50 per million output tokens — significantly cheaper than GPT-5-class models and Claude Sonnet 4.x at the same quality tier. The 256K context window is included at no premium, and Moonshot's automatic context caching can reduce input costs by up to 75% for repeated or overlapping prompts, which is particularly useful for long-document RAG and agent workflows.

Note: Always verify current API and subscription pricing on platform.moonshot.ai and kimi.com before committing to a plan. Pricing has shifted several times during the K2 → K2.5 transition.

Key Features

• 256K-token context window: Large enough to ingest a full novel, a 200+ page PDF stack, or a midsize codebase in a single conversation without splitting into chunks.
• Kimi K2.5 MoE model: ~1T total / ~32B active parameters using Mixture-of-Experts, giving strong reasoning and coding performance at low inference cost.
• Kimi K2 Thinking variant: A reasoning-tuned version of K2 for harder multi-step problems, benchmarks, and research-style tasks.
• Agent Swarm mode: Coordinates multiple specialized sub-agents in parallel on complex tasks, which Moonshot reports can meaningfully reduce wall-clock time versus sequential approaches.
• Automatic context caching: Repeated or overlapping prompts reuse cached context so input costs drop by up to 75% — particularly valuable for RAG and agent loops.
• Multi-file upload and analysis: Upload PDFs, Word documents, spreadsheets, and code files and ask Kimi to cross-reference and summarize them.
• Web search integration: Kimi can pull current information from the web to ground answers and citations.
• Deep Research mode: A longer-running research workflow where Kimi breaks a query into sub-questions, searches, and drafts a structured report.
• OK Computer agent: Moonshot's agent runtime that executes multi-step tasks on behalf of the user, with higher quotas on paid tiers.
• Kimi Code CLI: Terminal-native coding assistant for developers who prefer a command-line workflow.
• Bilingual Chinese–English support: Kimi is particularly strong in Chinese and offers solid English quality as well.

Best For

Cost-conscious developers building on LLMs: At roughly $0.60 / $2.50 per million input/output tokens and effective context caching, Kimi K2.5 is one of the cheapest capable long-context models on the market. Good pick for high-volume RAG, agent loops, and research pipelines.

Researchers and analysts processing long documents: The 256K context window comfortably handles entire books, long legal contracts, multi-file research projects, and large codebase reviews in a single conversation.

Bilingual and Chinese-market users: Kimi has first-class Chinese language support and is widely used inside China. If you're building for or working with Chinese-language content, Kimi is one of the strongest options.

Users who want to try agentic AI cheaply: Agent Swarm and OK Computer give you a low-cost way to experiment with multi-agent workflows, long-running tasks, and autonomous research without burning through frontier-model budgets.

Pros & Cons

FAQ

Score based on product quality, usability, value, features, reliability, integrations & market trust. How we score →