Kimi K3.1 Pro as a Labs experiment, open for a short test window: temporary, not a permanent id. Moonshot's newest open model, post-trained from Kimi K3 and served from the official Kimi-K3.1-Pro open weights: a 2.8T mixture-of-experts (104B active per token) with native image understanding and a 1M-token context, which Moonshot reports as substantially stronger than K3 on coding, factual accuracy and agentic work. It thinks by default at maximum reasoning effort; select none, low, high or max to trade depth for speed. Access is seat-gated through the Labs page while an experiment is live. It is offered at limited capacity and low availability, so expect it to be flaky and to go down under load: crash it, give it a moment, and try again. For production work we recommend the pay-per-token umans-kimi-k3.1-pro.
Umans Kimi K3.1 Pro (lab) Playground Retired
playground experiment; the Kimi K3.1 Pro lab ran Nov 4 to Nov 7, 2026; its metrics before the Nov 6 release carry on the released model's status page as its pre-release period
retired Nov 7, 2026 · replaced by umans-kimi-k3.1-pro · weights ↗
74.4tok/s
throughput · p50 · whole period
2.22s
TTFT · p50 · whole period
100.00%
uptime · whole period
Trends
Nov 4, 2026retired Nov 7, 2026Nov 9, 2026
Nov 4, 2026retired Nov 7, 2026Nov 9, 2026
Changelog
Nov 72026
Playground closed: Umans Kimi K3.1 Pro (lab) Testing
The Kimi K3.1 Pro lab closed on November 7, 2026. It opened November 4 and ran past the announced November 5 close. The model graduated on November 6 as the pay-per-token umans-kimi-k3.1-pro, and the lab's metrics up to that release live on its status page as the pre-release period. Thanks to everyone who tested it.
Nov 42026
New Labs experiment: Umans Kimi K3.1 Pro Testing
A new lab opened on umans-kimi-k3.1-pro-lab: Kimi-K3.1-Pro served from Moonshot's official open weights, the first model of the K3.1 series, post-trained from Kimi K3: a 2.8T mixture-of-experts (104B active per token) with native image understanding and a 1M context window, which Moonshot reports as substantially better at coding, factual accuracy and agentic work. It thinks by default at max effort; select none, low, high or max to trade depth for speed. Free and seat-gated while the experiment runs, served at limited capacity and low availability: it is a lab, so expect it to be flaky and to go down under load. Crash it, give it a moment, and try again. The window closes November 5, 2026 at 23:00 UTC (midnight in Paris).