XNews

Claude Opus 5.5 Release and Prompt Optimization

Claude Opus 5.5 Release and Prompt Optimization

Anthropic released Claude Opus 5.5, matching Fable 5.1 performance on most tasks while costing 40% less than Opus 5. It features lower token prices, 30% faster output, and improved capabilities in code review, knowledge work, and visual understanding. Available on Claude platform, AWS, Google Cloud, and Azure.

Key insights

Performance and Cost Improvements

Claude Opus 5.5 matches Fable 5.1 on most work tasks while costing 40% less than Opus 5. Input and output prices are $4 and $20 per million tokens, 20% cheaper than Opus 5. Cache read dropped from $0.50 to $0.20. Output speed increased by over 30%. In a HAProxy rewrite from C to Rust, Opus 5.5 completed in 9.5 hours vs Fable 5.1's 12 hours, with 51% lower cost.

Benchmark Results and Real-World Tests

On Terminal-Bench 4.0, Opus 5.5 scored equal to GPT-6 Astra at about 40% the cost. In a financial report generation test requiring no fabricated numbers or quotes, Opus 5.5 passed 16 out of 18 reports, while Fable 5.1 and Opus 5 passed none. An early tester audited and fixed a 200,000-line codebase in under 3 hours, compared to Opus 5's 20+ hours.

Safety Restrictions and Distillation Prevention

Opus 5.5's capabilities in cybersecurity and biology approach Mythos 5.1, so it inherits Fable 5.1's safety measures: most cybersecurity tasks are routed to older Opus 4.8, and biological research is restricted with a Life Sciences Verification Program for legitimate institutions. Thinking mode cannot be disabled. API accounts registered after August 31 cannot extract reasoning via history manipulation to prevent distillation.

Prompt Optimization Adjustments Required

A detailed guide from A\ outlines 11 adjustments for Opus 5.5. Key changes: effort calibration (5.5's medium ≥ 5's high), thinking mode always on, removal of instructions that write reasoning into text (use display: 'summarized'), and new handling for progress updates. The model is more direct in writing and better at following style instructions.

Improved Visual Understanding and Code Review

Opus 5.5 shows significantly better visual understanding: even at lowest effort, it reads dense charts more accurately than Opus 5 at highest effort, with fraction of token usage. It is more reliable for tasks where position carries semantics (flowcharts, calendar screenshots). In code review, it has fewer false positives and false negatives, and can sustain multi-hour unattended audits.

Subscription Benefits and Availability

Pro, Max, and Team subscribers get increased five-hour usage limits and an extra reset that can be saved. Opus 5.5 is available on Claude platform, AWS, Google Cloud, and Azure with API model name claude-opus-5-5. Sonnet 5.5 and Haiku 5.5 will follow in coming weeks.

Source posts