Kimi

Kimi is the AI assistant and model family from Moonshot AI. Moonshot sells API access to its models on the Kimi API platform, separate from the consumer Kimi membership and from Kimi Code, its coding product. The platform documentation now lists four lines. Kimi K3.1 Pro, launched October 23, 2026, is the flagship and the first K3.1 model, with a 1 million token context window and a faster Highspeed variant (API ids kimi-k3.1-pro and kimi-k3.1-pro-highspeed); Moonshot says its full weights will be released by November 4, 2026. The others are Kimi K3, Kimi K2.7 Code for programming work, and Kimi K2.6. The K2 models are on their way out: from October 31, 2026 they are no longer offered to newly registered API users (existing users keep access), and Moonshot plans to retire the whole K2 series by the end of December 2026, so new projects should start on K3 or K3.1.

API inference is billed pay as you go, with no subscription plan. Moonshot charges for input and output tokens separately, with prices quoted per million tokens and taxes added at checkout based on where you are. As a rough guide from the docs, one token is about three to four characters of typical English text, and an API endpoint reports the exact token count for each call.

The K3 series adds automatic context caching. When requests repeat the same prefix, the platform caches it with a lifetime of either five minutes, the default, or one hour. Writing a prefix into the cache is billed by the chosen lifetime, and any later input that hits the cache is billed only at the lower cached input price. Each hit refreshes the cache lifetime without another write charge. This matters for apps that send the same long system prompt or document many times.

Kimi has long focused on reading long documents. If you upload a file and pass the extracted text to the model, that text is billed as normal input tokens, but the file extraction and storage APIs themselves are free for now, according to the pricing page.

Because the consumer membership, Kimi Code and the API are billed differently, check which product you need before you pay. For comparisons with Western flagships such as ChatGPT, Claude, Gemini and Grok, see our assistant guides. Moonshot's live pricing tables are on its platform site, and they can change, so recheck before committing to a budget.