Moonshot Shuts Kimi K2 API to New Users

Moonshot AI stopped offering its Kimi K2 models to newly registered API users on Saturday, Oct. 31, and says it plans to retire the entire K2 series by the end of December 2026. Existing accounts keep access for now, according to a notice on the Kimi API platform's model list, which points developers to the newer kimi-k3.1-pro.

Key takeaways

• Since Oct. 31, new Kimi API sign-ups cannot use K2-series models; existing users "are not affected," Moonshot says.
• Moonshot "plan[s] to retire all K2 series models by the end of December 2026" and recommends kimi-k3.1-pro.
• K3.1 Pro is not a drop-in swap: thinking is always on, reasoning effort defaults to the maximum and sampling settings are fixed.

Which Kimi K2 models are affected?

The notice on Moonshot's Model List page reads: "starting October 31, 2026, K2 series models will no longer be available to newly registered users." It does not list model IDs.

Several K2 models are already in the page's Deprecated Models section, including kimi-k2-0905-preview, kimi-k2-0711-preview, kimi-k2-turbo-preview, kimi-k2-thinking and kimi-k2-thinking-turbo. The page says kimi-k2.5 was discontinued on Aug. 31, 2026.

Three K2-numbered models remain on the active list: kimi-k2.6, kimi-k2.7-code and kimi-k2.7-code-highspeed. Moonshot's pricing page groups "K2 series models" by their 262,144-token context window, but the company has not said model by model whether the December date applies to them. Developers on those IDs should treat the deadline as likely and confirm with Moonshot.

Kimi K3.1 Pro migration: what changes

According to Moonshot's K3.1 Pro quickstart, the API still uses the OpenAI SDK with the base URL api.moonshot.ai/v1, so most client code carries over. The behavior does not.

• Thinking cannot be turned off. The reasoning_effort field accepts low, high or max, and the default is max. Moonshot suggests low to cut latency.
• Sampling is fixed. Temperature 1.0, top_p 0.95, n=1 and zero presence and frequency penalties are locked, and Moonshot says to omit them.
• Longer outputs by default. max_completion_tokens defaults to 131,072 and can rise to 1,048,576, matching the K3 series' 1,048,576-token context window, four times K2's 262,144.
• Images change. Public image URLs are not supported; images go in as base64 or an uploaded file ID.
• Multi-turn and tools. Apps must send back the complete assistant message, not only its content field.

Kimi API price comparison

Moonshot's own pricing page did not render prices when AI Newsway checked it. Third-party tracker BenchLM, updated Oct. 29, lists these rates per million tokens. Moonshot says batch requests for K3.1 Pro and K3 cost 60% of standard.

The gap widens in practice because K3.1 Pro always produces reasoning tokens, and at max effort by default.

Context

Moonshot released K3.1 Pro in late October and has promised full open weights by Nov. 4. The retirement follows a year in which Chinese open-weight labs have shipped new generations quickly, as when Qwen3.8 Max climbed above every American lab but two on one independent index.

What's next

The two dates to watch are Nov. 4, for the K3.1 Pro weights, and the end of December, when K2 access is slated to end. Moonshot has not published an exact shutdown day.

FAQ

I already use Kimi K2. Does my app stop working now?

No. Moonshot says existing users are not affected by the Oct. 31 change. It plans to retire all K2 models by the end of December 2026.

What should I migrate to?

Moonshot recommends kimi-k3.1-pro. A kimi-k3.1-pro-highspeed variant, which Moonshot says runs two to three times faster, and kimi-k3 are also available.

References

• Model List — Moonshot AI Kimi API Platform (accessed Nov. 1, 2026)
• Kimi K3.1 Pro Quickstart — Moonshot AI Kimi API Platform (accessed Nov. 1, 2026)
• Kimi API Pricing (October 2026) — BenchLM (Oct. 29, 2026)