Moonshot AI releases Kimi K3.1 Pro model with 3 trillion parameters
Moonshot AI announced the Kimi K3.1 Pro on October 24 2026, a 3 trillion‑parameter model that supports up to 1 million‑token contexts for certain plans. The company also introduced a high‑speed variant that is 2–3× faster than the regular version and is available on Kimi Web, Work, Code, and API for Pro and higher tiers.
Key points
- Moonshot AI announces Kimi K3.1 Pro, a 3 trillion parameter model with 1M token context support.
- High‑speed version offers 2–3× faster response, available on Kimi Web, Work, Code, and API for Pro and higher plans.
- Internal xdev‑bench‑hard pass@3 scores 81.0% for K3.1 Pro vs 90.5% uncontrolled, 33.3% for previous K3.
The model’s release includes a 3D Make feature that turns text into 3D models, useful for design drafts but not guaranteed to be manufacturing‑ready. Moonshot AI said it intentionally curtailed dangerous capabilities by filtering cyber‑related content during training and reinforcement learning. In its internal xdev‑bench‑hard test, K3.1 Pro scored 81.0% pass@3, compared with 90.5% for an uncontrolled version and 33.3% for the previous K3.
Developers are cautioned to avoid deploying code or designs generated by the model directly into production or manufacturing. Weight files are slated for release on November 4 2026, but the open‑weight version is not yet complete.
The story so far
2 episodes →- Moonshot AI releases Kimi K3.1 Pro model with 3 trillion parametersthis story
Deliberately 'Curbing' AI Capabilities? The Challenges for OpenAI Seen in Kimi K3.1 Pro
note.com · 26 October 2026
Loading the full article…
This text was published by note.com and written by nagomi97. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- sFNX runs Qwen3.8-Flash-Next at 382 tokens/s using three used V100 GPUs · 1 src
- OpenAI announces GPT-6.1 Sol with claimed 86.6% to 97.9% COBOL migration accuracy · 8 src
- CodeRabbit uses GPT‑6.1 Sol in security agent · 1 src
- Anthropic reports 3.6x jump in token volume after Claude Opus 5.5 launch · 3 src
- AI models may struggle with clear communication, study finds · 2 src
Comments
via GitHub Discussions