Top-ranked DeepSeek V4 Flash stumbles on real agent tasks
DeepSeek's leaderboard-topping V4 Flash completed only 53.8% of complex agent tasks in testing. Thirty hard tasks were run across eight different agent harnesses.
Models
AI agents
DeepSeek made three moves at once: model, open source, price rise
DeepSeek took V4-Pro out of testing, open-sourced its agent software and raised API prices. The new rates make usage outside Chinese business hours cheaper.
Models
A 3-billion-parameter agent closes in on models ten times its size
Microsoft Research's open-source Orchard framework provides shared infrastructure for agent training. Orchard-SWE scores 69.7 percent on SWE-bench Verified.
Models
NVIDIA releases a 30-billion-parameter open model for agents
NVIDIA has released Nemotron 3.5 Lightning for long-running agentic workloads, alongside NeMo Switchyard, an open source library that routes each request to the most suitable model.
Models
Grok 4.6 lands: not a bigger model, a longer training run
SpaceXAI's new model invests in post-training rather than a larger base. Grok 4.6 brings a 500K-token context, an agent focus and API pricing that undercuts rivals.
Models
AI agents given real research questions were rejected by reviewers
Princeton and the UK AI Security Institute handed AI agents the research questions from two unpublished NeurIPS papers. Both resulting papers were rejected.
Research
DeepSeek Unveils Open-Source Harness to Rival Claude Code
DeepSeek has announced DeepSeek Harness, an open-source agent framework for developers, alongside its updated flagship model V4-Pro; meanwhile, API prices are set to rise.
Products
Hugging Face Reproduces 2,226 ICML Papers Using AI Agents
In a hackathon organized by Hugging Face, 1,221 participants used coding agents to attempt to reproduce 2,226 papers accepted at ICML 2026.
Research