Claude Opus 5 goes live with a 1M-token window and $5 input pricing
数字生命卡兹克 · wechat · 2026-07-25
Claude Opus 5 lands with stronger agent performance and cheaper pricing
This long Chinese post says Claude Opus 5 has gone live across Claude products, with a 1M-token context window and a knowledge cutoff of May 2026.
Key takeaways from the author’s hands-on use:
- Opus 5 is positioned as a strong agentic executor, especially good at using tools, checking its own work, and retrying
- The post claims it performs very well on hard engineering and agent tasks, while Claude Fable 5 still feels better for planning and high-level design
- Pricing is listed at $5 / million input tokens and $25 / million output tokens, with a Fast Mode that is about 2.5× faster at double the price
The article also argues that Claude Code is moving toward a lighter workflow: fewer rules, fewer redundant skills, more environment design, and more judgment delegated to the model itself.
The author finishes with a practical stack for different tasks, including Fable 5 for deep planning, Opus 5 for design/review and bug fixing, GPT-5.6 Sol for some execution and web-based review, and Kimi K3 for frontend demos.
More from coding & agent
- A Type Proposal Would Make Inline React Arrays Fail at Build Time — aidenybai · 2026-07-25
- Long-running agents need trustworthy verification, not just more loops — hugobowne · 2026-07-25
- A sarcastic warning that AI may just create more monorepo per capita — HanchungLee · 2026-07-25
- Linear adds Claude Opus 5 to Coding Sessions for drafting PRs — generativist · 2026-07-25
- Opus 5 plus Unity CLI feels like a cheat code for vibe-coding games — chongdashu · 2026-07-25
- OpenRouter launches beta classifiers to tag and visualize model routing — AccBalanced · 2026-07-25