Claude Sonnet 5
Points and comments are a snapshot, not live.
Anthropic releases Claude Sonnet 5, its most capable agentic Sonnet model yet.
Claude Sonnet 5 is available across all plans starting today at introductory pricing of $2/M input tokens and $10/M output tokens through August 31, 2026 (then $3/$15). It narrows the gap with Opus 4.8 on agentic tasks like reasoning, tool use, and coding, while being cheaper. Safety evaluations show lower rates of undesirable behaviors than Sonnet 4.6, though higher misaligned behavior rates than Opus 4.8. It has substantially poorer cybersecurity exploit capabilities than Opus 4.8, and ships with real-time cyber safeguards enabled. Early access partners report it finishes complex multi-step tasks where previous Sonnets stalled, at an attractive price point.
What commenters are saying
Several commenters note the agentic search chart shows Sonnet 5 at high effort costing more than Opus 4.8 with worse performance, making Opus the better choice for hard tasks. Some argue rapid improvement is plateauing and that faster models matter more than smarter ones now. Others find Sonnet 5 a strong incremental upgrade for well-scoped tasks when paired with Opus for planning. A few compare Sonnet 5 unfavorably to open-weight models on price/performance, citing GLM 5.2. The desire for the return of Fable and a new Haiku model is also expressed.