XNews

Claude Opus 5.5 Released with Improved Performance and Lower Cost

Claude Opus 5.5 Released with Improved Performance and Lower Cost

Anthropic released Claude Opus 5.5, matching Fable 5.1 on most tasks while costing 40% less. Input/output prices are reduced by 20%, cache reads by 60%, and output speed is up 30%. In a code audit of a 200,000-line codebase, Opus 5.5 completed the task in under 3 hours versus 20+ hours for Opus 5, using 2.5x fewer tokens.

Key insights

Opus 5.5 Matches Fable 5.1 Performance at Lower Cost

Anthropic stated that Claude Opus 5.5 performs on par with its Mythos-tier Fable 5.1 on most tasks, while reducing overall cost by 40%. The cost savings come from both lower per-token prices and reduced token usage. Input and output prices are $4 and $20 per million tokens respectively, 20% lower than Opus 5. Cache read price dropped from $0.50 to $0.20 per million tokens, a 60% reduction that benefits heavy agent users.

Real-World Code Audit Shows 7x Speed Improvement

An early tester used Opus 5.5 to audit and fix a 200,000-line codebase in under 3 hours, while Opus 5 took over 20 hours and used 2.5x more tokens. Internally, Anthropic had Opus 5.5 and Fable 5.1 rewrite the HAProxy load balancer from C to Rust. Opus 5.5 completed in 9.5 hours at 51% lower cost, while Fable 5.1 took 12 hours. Both passed nearly all regression tests.

Improved Factual Accuracy in Research Reports

In a test where models had to write a quarterly earnings report using only web search results, Opus 5.5 produced 16 out of 18 reports with no fabricated numbers or quotes. Fable 5.1 and Opus 5 failed all 18 reports. This indicates a significant improvement in factual grounding and adherence to source material.

LEGO Design Capability Showcased with Real-World Build

HuggingFace team member Victor Mustar used Opus 5.5 to design a Microduck robot using 1,113 real LEGO pieces in 6 colors. The model produced a 141-page, 237-step building manual with zero collisions, correct assembly direction, and estimated center of mass within the support polygon. It also generated LDraw files, CSV parts list, and BrickLink purchase XML, demonstrating advanced spatial reasoning and design skills.

Safety Restrictions Applied for Cybersecurity and Biology

Opus 5.5's capabilities in cybersecurity and biology are close to Mythos 5.1, so it inherits Fable 5.1's safety guardrails. Most cybersecurity tasks are automatically routed to older Opus 4.8. Biological research is also restricted, with a Life Sciences Verification Program available for academic labs and pharmaceutical companies to request access. A verification channel for security practitioners will open in coming weeks.

API Changes: Thinking Mode Mandatory, Distillation Protections Added

Developers should note that thinking mode can no longer be disabled. Additionally, API accounts registered after August 31, 2026 cannot extract the model's reasoning process by modifying historical context, a measure to prevent distillation attacks. Existing integrations should review the documentation. Opus 5.5 is available on Claude platforms, AWS, Google Cloud, and Azure as claude-opus-5-5.

Source posts