Anthropic releases Claude Opus 5, its most capable and most aligned model at half the price

Anthropic released Claude Opus 5, describing it as its most capable Opus model and least susceptible to being tricked into misuse. The model approaches Claude Fable 5 capabilities at roughly half the price, delivering the company's strongest work performance with notable improvements in coding, science, autonomy, self-correction, and error recovery. It's now the default model on Claude Max and the strongest option on Claude Pro, with new developer features including mid-conversation tool changes and automatic API fallbacks.

On AWS, Opus 5 arrived on Amazon Bedrock with zero data retention compatibility. AWS's Swami Sivasubramanian emphasized its agentic strengths: long-running agents that 'work for hours and even overnight,' navigate codebases like an experienced engineer, recover from errors, and push back on flawed instructions rather than executing blindly, breaking complex jobs into sub-agents needing less supervision.

Benchmark reactions were strong. François Chollet noted Opus 5 set a new state of the art on ARC-AGI-3 at 30% — a benchmark measuring problem-solving with no prior exposure, where scaling has historically bought the least — calling it an 'impressive jump.' Anthropic's Boris Cherny highlighted that Opus 5 is the company's 'least prompt injectable model yet,' a claim that landed pointedly given OpenAI's sandbox-escape disclosure the same week.

The official Anthropic announcement dominated Hacker News (1,346 pts, 730 comments), making it the day's top model-release story. Skeptics offered caveats: LlamaIndex's Jerry Liu benchmarked Opus 5 on document parsing via ParseBench and found it only 'roughly on par with Opus 4.8' and average at OCR — advising against using it to parse documents at scale at 8¢ per page. Watch for pricing pressure against DeepSeek V4 and Alibaba's Qwen.