---

title: “Claude Opus 4.8 Release” topic: “Claude Opus 4.8 Core Updates” lang: “ZH” —

🚀 Core Update: Claude Opus 4.8

1. Key Performance Enhancements

• Agentic Capabilities: Sharper judgment in agentic tasks; error self-detection improved (code flaw miss rate reduced by 4x compared to predecessor).
• Benchmarks:
  • Super-Agent Benchmark: Only model to complete all cases end-to-end, beating GPT-5.5 at cost parity.
  • Online-Mind2Web: Scored 84%, a significant jump over Opus 4.7 and GPT-5.5.
  • Legal Agent Benchmark: Record high score; first to break 10% on the all-pass standard.

2. New Feature Set

• Dynamic Workflows: [Claude Code Only] Ability to plan and run hundreds of parallel sub-agents in a single session for codebase-scale migrations.
• Effort Control: Users can now select response depth on claude.ai (Low Effort $ ightarrow$ Faster/Saves quota; High Effort $ ightarrow$ Deeper reasoning).
• Messages API Upgrade: Now accepts system entries inside the messages array, allowing instruction updates without breaking prompt cache.
• Fast Mode: 2.5x speed increase; 3x cheaper than previous models.

3. Pricing & Availability

• Standard Mode: Input $5/M tokens $\mid$ Output $25/M tokens.
• Fast Mode: Input $10/M tokens $\mid$ Output $50/M tokens.
• Availability: Available now via Claude API.

4. Roadmap

• Project Glasswing: Upcoming higher-intelligence models (currently in limited preview for cybersecurity).