Model
kimi-k3
Appears alongside
Featured in threads
Tracks
- Models & capabilities 5
- Open weights & ecosystem 4
- Safety & alignment 3
- Security & misuse 2
- Benchmarks & progress 2
- Government & policy 1
- Ideas & essays 1
UK AISI studies find safeguards on biological AI tools can be undone or bypassed
In one study, fine-tuning restored viral-protein performance for about $80 of H100 time; in the other, agents rebuilt simple bio tools in as little as four hours for $4 to $20.
Safety & alignment · Government & policy
Moonshot releases Kimi K3.1 Pro and says it removed cyber data before an open release
Priced at $2 and $10 per million tokens; weights followed by 4 November, with model-as-a-service operators above $20m in revenue required to sign a separate agreement.
Models & capabilities · Open weights & ecosystem · Security & misuse
A Goodfire researcher argues neural networks are far more interpretable than the field believed
Nick Cammarata estimates about 90% of Qwen3-32B's neurons have one main job, and says the 2020 superposition evidence he co-authored was misread; a companion paper is still to come.
Safety & alignment · Ideas & essays
Study detects reward hacking from models' internal representations
Cheap difference-of-means probes on frontier open-weight models matched expensive LLM-judge monitors at catching reward hacking, at a fraction of the compute cost.
Safety & alignment · Benchmarks & progress
Mindgard discloses a jailbreak of Moonshot's Kimi models
The security firm said Moonshot did not respond to its private report for nearly two months, until the BBC sought comment ahead of publication.
Security & misuse
Harvey builds its first in-house model on Moonshot's Kimi K3
Named Tenet, it is post-trained from a Chinese open-weight base rather than a closed US model, and Harvey reported near-double the task-completion rate of stock Kimi K3.
Models & capabilities · Open weights & ecosystem
AI models score perfect marks at International Mathematical Olympiad 2026
Only two of the six perfect scores came from official IMO graders; the other four were self-administered and graded by a Claude-based agent rather than human judges.
Benchmarks & progress · Models & capabilities
Moonshot AI launches Kimi K3
A mixture-of-experts design activating 104 billion of its 2.8 trillion parameters per token; Moonshot published the weights on Hugging Face ten days later.
Open weights & ecosystem · Models & capabilities
Moonshot AI releases Kimi K2.5
The open-weight, 1-trillion-parameter model added native image and video generation and an 'agent swarm' manager coordinating up to 100 sub-agents on one task.
Open weights & ecosystem · Models & capabilities