AnalysisAI ModelsJuly 28, 2026
User criticizes Opus 5 as extremely RL-fried and mistake-prone
A Reddit user reports that Anthropic's Opus 5 performs poorly on all tasks in a large project, including as a subagent, calling it "extremely RL-fried and mistake-prone."
1 source
More stories today
- GM triples merged pull requests with AI agent workflows
- Runway turns AI video avatar drift bug into feature
- NVIDIA Jetson AI Lab shows local GenAI with Ollama, vLLM, llama.cpp
- AI agents fail silently with no error logs or alerts
- Discovering cryptographic weaknesses with Claude