[AINews] Kimi 3.1 Pro 104BA-2.8T: Fable-5.1-Class Open Weights narrows Closed-AI lead to 1 month... and Pro Highspeed API with 2-3× speed

Moonshot has released its new Kimi K3.1 Pro model, an upgraded version of the earlier K3 that adds further training and a mixture‑of‑experts architecture. The model contains roughly 2.8 trillion total parameters, of which 104 billion are active during inference, and can handle contexts of up to one million tokens. Its highlighted use cases include full‑code generation, computer‑aided design environments, financial modeling, and long‑duration agent execution.

The offering comes in two variants: a standard version and a “Highspeed” edition that promises responses two to three times faster, albeit at double the cost. The launch also introduces “Kimi Hosted Agents,” a 24‑hour managed environment for running extended tasks and coordinating multiple agents. In independent tests, the K3.1 Pro ranked fourth on Artificial Analysis’s Intelligence Index with a score of 55, and improved its standing in Code Arena, reaching eighth place in the WebDev category with 1,692 points. In a cybersecurity benchmark, it achieved 64 % on the Cyber Index, the second‑best public result.

This development is significant because it demonstrates that large‑scale models continue to grow in capability and speed, broadening the range of tasks that can be automated and providing new tools for developers and professionals who rely on AI for complex work. The forthcoming release of open weights in November will enable the community to study and adapt the model without depending on proprietary versions.

Read the original article on Latent Space

This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.