Anthropic Unveils Claude Opus 5.5: Delivers Fable 5.1-Grade Performance While Cutting Operating Costs by 40% Compared to Opus 5

Anthropic has unveiled Claude Opus 5.5, the debut model of its new Claude 5.5 family. According to the company, the model matches the performance of Fable 5.1 across most tasks while running 40% cheaper than Opus 5 on typical workloads at default settings. On Anthropic's internal benchmarks, it takes the top spot in agentic coding, computer use, and knowledge work, making it a significant release for teams deploying frontier AI in production.

Availability and API Access

Claude Opus 5.5 ships as a managed API model only; Anthropic is not releasing the weights, so self-hosting is off the table. Developers can access it through the model identifier claude-opus-5-5 on the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure. As with earlier Opus releases, zero data retention is available for customers who need it.

Benchmarks: A Strong Lead, With Caveats

Anthropic's reported scores use adaptive thinking at max effort with production safeguards enabled:

| Benchmark | Opus 5.5 | Fable 5.1 | Opus 5 | GPT-6 Astra |
|---|---|---|---|---|
| Terminal-Bench 4.0 | 66.4% | 55.8% | 52.3% | 57.9% |
| FrontierCode v1.1 | 54.4% | 50.3% | 48.0% | 53.3% |
| CursorBench 4.0 | 57.8% | 51.8% | 46.6% | n/r |
| GDPval-AA v2.1 (Elo) | 1846 | 1735 | 1708 | 1542 |
| OSWorld 2.0 | 81.8% | 80.7% | 74.0% | n/r |
| Terminal-Bench-Science 0.1 | 58.7% | 52.6% | 29.0% | 64.6% |
| AutomationBench | 40.0% | 31.4% | 26.9% | 41.4% |

Terminal-Bench 4.0 is reported at xhigh effort. GPT-6 Astra still leads on Terminal-Bench-Science and AutomationBench, and Zapier ran AutomationBench without fallback models, so safeguard interventions counted as failures. Anthropic itself cautions that benchmark margins are becoming less reliable as a guide, noting the gap to Fable 5.1 in its own internal use is narrower than the numbers suggest.

The cost-adjusted picture is more striking. At default medium effort, Opus 5.5 scores 54.6% on FrontierCode — above GPT-6 Astra's best of 53.3% at roughly a fifth of the cost per task. On CursorBench, medium effort reaches 52.5%, eleven points higher than GPT-5.6 Sol's best, at about a third of the cost.

Pricing and Speed

Because Opus 5.5 requires less compute to serve than Opus 5, prices fall accordingly:

| Per 1M tokens | Opus 5.5 | Opus 5 |
|---|---|---|
| Input | $4 | $5 |
| Output | $20 | $25 |
| Cache reads | $0.20 | $0.50 |
| Cache writes | $5 | $6.25 |

Cache reads, which dominate agentic and coding costs, drop 60%. Combined with lower token usage per task, this adds up to the headline 40% savings. Output generation is more than 30% faster than Opus 5, and a fast mode in Claude Code and the Claude Platform delivers up to 2.5x speed at $8 input and $40 output per million tokens.

What Early Testers Found

Feedback from early users highlights the efficiency gains. One tester completed a 680,000-line code migration in under a day, while another audited and repaired a 200,000-line codebase in under 3 hours — a task that took Opus 5 more than 20 hours and 2.5x the tokens. In an internal C-to-Rust port of HAProxy, Opus 5.5 finished in 9.5 hours versus 12 for Fable 5.1, at 51% lower cost. Deloitte reported that Opus 5.5 at lowest effort caught 72% of known review bugs, compared with 56% for Opus 5 at high effort. In a difficult earnings-report test, 16 of 18 Opus 5.5 outputs cleared Anthropic's quality bar; neither Fable 5.1 nor Opus 5 ever did. Anthropic also notes a shift in writing style: key information comes first, jargon is reduced, and user style instructions are followed more closely.

Safety, Safeguards, and API Changes

This is Anthropic's first release since CEO Dario Amodei called for pacing the frontier. External evaluators including METR and Frontier Design tested the model before launch, and it achieves the best score to date on Anthropic's automated behavioral audit spanning nearly 2,000 scenarios. In a new containment test, attempts to circumvent boundaries fell roughly 85% versus Opus 5; Anthropic adds that the model often suspects it is being evaluated. Biology and cyber capabilities are comparable to Claude Mythos 5.1, so Fable 5.1-class safeguards apply: routine cybersecurity bug finding is allowed while most other cyber tasks route to Opus 4.8, the Cyber Verification Program will expand to Opus 5.5, and vetted organizations can apply to the Life Sciences Verification Program. Preserved thinking prevents API users from editing prior context to extract reasoning, applied to accounts created on or after August 31, 2026. Thinking can no longer be disabled, and outputs carry watermarking for EU AI Act compliance. Full details appear in the Opus 5.5 System Card.

Key Takeaways

• Opus 5.5 matches Fable 5.1 on most work and beats both Opus 5 and Fable 5.1 on nearly every reported benchmark.
• API pricing drops to $4/$20 per 1M tokens, with cache reads falling 60% to $0.20.
• Anthropic estimates 40% savings on typical workloads, with output more than 30% faster than Opus 5.
• Cyber and biology requests face Fable 5.1-class safeguards, and thinking cannot be switched off.
• Weights remain closed; claude-opus-5-5 runs via Claude Platform, AWS, Google Cloud, and Azure.

The combination of Fable 5.1-level capability at significantly lower cost could shift how teams choose between frontier and mid-tier models for agentic workloads. Watch for whether competitors respond with their own price cuts, and how the expanded verification programs shape access to the model's restricted capabilities.

Meta description: Anthropic releases Claude Opus 5.5, matching Fable 5.1 performance while cutting running costs 40% versus Opus 5, with new pricing and safeguards.

Tags: Anthropic, Claude Opus 5.5, AI benchmarks, agentic coding, LLM pricing

Featured image: Abstract technology style visualization of layered neural network structures in cool blue and violet tones, no people, logos, or text.