Kimi✓ ReviewedAnnounced 16 July 2026Reviewed 22 October 2026Kimi K3 wants your biggest files and longest jobs. The benchmark claims still need your task.
Reviewed editorial Update
Kimi K3 wants your biggest files and longest jobs. The benchmark claims still need your task.
Kimi K3 combines published model weights, native visual input and up to one million tokens of context. Kimi Code documents low, high and max thinking effort. Moonshot lists K2.5 and the moonshot-v1 series as discontinued, so existing integrations need an available supported route.
What changed
Moonshot released Kimi K3 on 16 July 2026. Its model documentation describes 2.8 trillion parameters, native visual input and up to one million tokens of context; Kimi Code documents low, high and max thinking effort. The model weights are available under the Kimi K3 License. The platform model list records kimi-k2.5 and the moonshot-v1 series as discontinued on 31 August 2026 and the kimi-k2 series on 25 May 2026, with kimi-k3 named as a replacement. It does not specify a universal error code for calls to retired models.
Why it matters
Large context and published weights can make K3 worth evaluating for source-heavy work, but neither proves that it will preserve the details your task depends on. The older models are already listed as discontinued. A migration comparison must use saved evaluation cases and an available supported baseline, rather than depend on a retired endpoint still answering.
Bit’s takeaway
The human take
A person chooses the authoritative sources, checks the decision-bearing details, and owns the move to a supported service.
Turn this into a workflow
Bring your task to Bit.
Bit reevaluates the recorded facts when you ask. Historical recommendations do not approve a current action.
Ask Bit about my taskSource-backed facts
- Moonshot released Kimi K3 on 16 July 2026 and published model weights under the Kimi K3 License. Its documentation describes native visual input, up to a one-million-token context window and low, high and max thinking-effort controls in Kimi Code.
- Kimi (Moonshot AI) technical blog ↗
- Kimi (Moonshot AI) platform documentation ↗
- Moonshot: Kimi Code release notes ↗
- Moonshot: Kimi K3 model card and weights ↗
ReviewedTechnical record
[kimi-k3-fact; official_fact; product_fact; reviewed 2026-10-22; sources: Kimi (Moonshot AI) technical blog (https://www.kimi.com/blog/kimi-k3); Kimi (Moonshot AI) platform documentation (https://platform.kimi.ai/docs/models); Moonshot: Kimi Code release notes (https://www.kimi.com/code/docs/en/kimi-code/whats-new.html); Moonshot: Kimi K3 model card and weights (https://huggingface.co/moonshotai/Kimi-K3)] Moonshot released Kimi K3 on 16 July 2026 and published model weights under the Kimi K3 License. Its documentation describes native visual input, up to a one-million-token context window and low, high and max thinking-effort controls in Kimi Code.
- Moonshot discontinued kimi-k2.5 and the moonshot-v1 series on 31 August 2026 as scheduled, and its model documentation records them as no longer maintained or supported, with kimi-k3 named as the replacement. The kimi-k2 series was discontinued on 25 May 2026.Reviewed
Technical record
[kimi-legacy-retirement-fact; official_fact; product_fact; reviewed 2026-10-18; sources: Kimi (Moonshot AI) platform documentation (https://platform.kimi.ai/docs/models)] Moonshot discontinued kimi-k2.5 and the moonshot-v1 series on 31 August 2026 as scheduled, and its model documentation records them as no longer maintained or supported, with kimi-k3 named as the replacement. The kimi-k2 series was discontinued on 25 May 2026.
Sources and method