Moonshot AI says it has released Kimi K3.1 Pro and plans to make the complete weights available before November 4
BIBIBI
AT A GLANCE
Moonshot AI releases Kimi K3.1 Pro and a high-speed version, and plans to make the complete weights available before November 4.
Article
Moonshot AI releases Kimi K3.1 Pro; model weights will be open-sourced before November 4 PANews October 24 reports that, according to Moonshot AI's official blog, Moonshot AI has released K3.1 the first model in the series, Kimi K3.1 Pro, and the high-speed version Kimi K3.1 Pro Highspeed, the latter operating at 2 to 3 times the speed of the standard version. Both models support 1000000-token context. Official data show that K3.1 Pro scored Terminal-Bench 4.0 in 65.70%, higher than GPT-6 Astra's 58.20%, and slightly lower than Claude Opus 5.5's 66.40%; in internal hallucination testing, the hallucination rate for single statements decreased from K3's 23.40%to 8.30%. The new models are now available on Kimi.ai, Kimi Work, Kimi Code, and API, with the complete model weights planned to be made available before November 4.
Moonshot AI releases Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed。
02
Kimi K3.1 Pro Highspeed operates at 2 to 3 times the speed of the standard version.
03
Both models support 1000000-token context.
04
Official data show that K3.1 Pro scored Terminal-Bench 4.0 in 65.70%。
05
K3.1 Pro is now available on Kimi.ai, Kimi Work, Kimi Code, and API。
06
The complete model weights are planned to be made available before November 4.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
This is a newly released model series comprising a standard version and a high-speed version. The report also mentions its test results, context length, launch channels, and planned release date for the complete weights.
Why it matters to readers
If the weights are released as planned, developers may gain more direct access to use and research the model; however, the release timing and specific license still require verification.
Beginners can first focus on the speed difference between the standard and high-speed versions, the context length, and whether clear access methods and usage licenses will be available after the weights are released.
Risks and unknowns
The evidence states that the information comes from Moonshot AI's official blog, but does not provide a link to the original official blog post.
The evidence publication date is October 23, while the text is labeled “October 24 report,” creating a timing inconsistency.
Whether the complete model weights will actually be made available before November 4 has not yet been confirmed.
The weights' license, access channels, and permitted scope of commercial use are not specified.
Terminal-Bench 4.0 The testing conditions for the internal hallucination test are not specified.
Related Developments
Loading event timeline…
Related concepts
-token context
This term is not in the glossary yet. Browse related concepts in the glossary.
