AIModel release24h ago5 sources
Loading…
The feed · Sunday 25 October
AI 5
Models, labs and the companies building them
★ Model releases
#1AIModel release24h ago5 sources
Moonshot AI released Kimi K3.1 Pro, a 3‑trillion‑parameter open‑weight model with a 1‑million‑token context window and a high‑speed variant that is 2‑3× faster. The model scores 55 on the Artificial Analysis Intelligence Index, an 11‑point gain that places it fourth behind three Anthropic and Meta models, and it nearly matches Claude Opus 5.5 on a coding test. Moonshot will publish the model’s weights by 4 November 2026 and has removed hacking‑related training data to limit misuse.
Why it matters: Developers can now run a top‑tier, open‑weight LLM locally and customize it, with full weights becoming available soon.
#2AI2h ago
Microsoft Research released the Agensh harness as open source. The system operates without a central orchestrator, letting self‑organizing workers share a Git repository, chat, and a findings board. It can scale from a single agent to 1,024 agents on ProgramBench within a six‑hour budget.
Why it matters: It provides developers a ready‑to‑use framework for building large‑scale, decentralized multi‑agent applications.
AI
MarkTechPost
#3AI2h ago
Cognition released the Devin Vulnerability Fixes plugin, which includes three MIT‑licensed agent skills. The skills inventory, patch, and test PHP, Node.js, and self‑managed GitLab deployments affected by recent critical RCE flaws, and automatically open one pull request per repository. The plugin is offered as an open‑source agent plugin.
Why it matters: It lets security teams automate remediation of high‑severity vulnerabilities across common development stacks.
AIMarkTechPost
#4AI14h ago
OpenAI fired safety researchers Tomek Korbak, Mikita Balesni and Jasmine Wang, who publicly contest the company's claim that they leaked confidential information. The trio says they were never provided in writing with details of the alleged mishandled data and link the accusations to their advocacy for more scrutinizable AI systems.
Why it matters: The dispute highlights challenges around transparency and accountability in AI safety teams, affecting how tech firms manage internal research and external scrutiny.
#5AI15h ago
A three‑month field experiment with 133 patent attorneys found AI assistance improved immediate performance, especially for junior lawyers. When AI was withdrawn, only senior attorneys maintained the gains, indicating baseline expertise is needed for lasting skill development.
Why it matters: Understanding AI's short‑term impact and limits helps legal tech teams design tools that foster sustainable skill growth.
