Claude Opus 5.5 delivers Fable 5.1-level performance at a reduced price while cutting back on its signature "Claudish" prose

Anthropic has released Claude Opus 5.5, the first model in a new generation of its Claude lineup. The company claims the new Opus delivers performance on par with Claude Fable 5.1 while running faster and at a significantly lower cost than the previous Opus, a move aimed directly at customer complaints about pricing and efficiency.

Lower token prices and reduced usage drive down operating costs

Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens, compared with $5 and $25 for Opus 5 — a 20 percent reduction. Cache read costs have also been cut by 60 percent.

According to Anthropic, total operating costs, factoring in both token prices and how many tokens the model consumes, will run roughly 40 percent lower than Opus 5's. The model also generates output more than 30 percent faster.

Anthropic leans on coding benchmarks to back up its pricing argument. On FrontierCode, the company says Opus 5.5 outperforms OpenAI's GPT-6 Astra at roughly 20 percent of the cost per task. On Terminal-Bench 4.0, it claims equal performance with Astra at 40 percent of the cost, and on CursorBench it says the new model beats GPT-5.6 Sol by 11 points at one-third the cost. The price reductions reflect mounting pressure from OpenAI and, in particular, Chinese AI models that offer somewhat lower performance at a fraction of the price.

Anthropic plans to follow up with Claude Sonnet 5.5 and Haiku 5.5 in the coming weeks, promising similar improvements in performance, efficiency, and safety. Its benchmarks place the new Opus ahead of both Fable 5.1 and the considerably pricier GPT-6 Astra on most tasks.

Less "Claudish" writing and built-in safeguards

Anthropic says Opus 5.5 communicates more naturally than earlier versions, leading with the most important information, avoiding jargon, and adhering more closely to writing instructions. Early testers found its output clearer and easier to follow, making it better suited to extended work sessions. Current Claude models have faced repeated criticism for formulaic, convoluted prose — a style often dubbed "Claudish."

Opus 5.5 is also the first Opus model to carry safeguards covering cybersecurity, biology, and frontier LLM development at the same level as Fable 5.1. When these safeguards trigger, requests are transparently routed to other models: users can still find and fix bugs in their own code, but most cybersecurity tasks shift to Opus 4.8, while requests flagged for biology or frontier LLM development go to Opus 5.

Verified organizations can apply to use the model for biological research through the Life Sciences Verification Program, and Anthropic intends to extend its existing Cyber Verification Program to Opus 5.5 in the coming weeks. The model is available now across all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure, with developers on the Claude Platform able to access it via the model ID claude-opus-5-5.

Stricter screening, anti-distillation measures, and EU compliance

Looking ahead, Anthropic says it plans to screen reinforcement learning environments more rigorously, arguing that flawed training environments are a major source of misaligned behavior. It is also developing improved alignment rewards, automated generation of safety training scenarios, and stronger monitoring.

The company is joining a broader industry debate over release timelines: OpenAI has recently called for international standards covering self-improving AI systems, and several researchers have publicly urged labs to slow down until alignment methods mature. Anthropic is pushing for a higher safety bar for models capable of fully automating AI research, and says public policy should play a larger role in defining it. External organizations Frontier Design and METR evaluated Opus 5.5 ahead of release.

Anthropic is additionally rolling out defenses against what it terms distillation attacks, in which adversaries use thousands of fake accounts to extract a model's capabilities at industrial scale and build powerful models without its safeguards. It cites a September 2026 threat report detailing distillation activity it has detected and blocked.

Opus 5.5 ships with "Preserved Thinking," an anti-distillation feature first introduced with Fable 5.1 that prevents API users from altering Claude's prior context to harvest its reasoning. It applies to Fable 5.1 and Opus 5.5 for API accounts created on or after August 31, 2026. The model also includes watermarking to comply with the EU AI Act, can no longer run with Thinking mode disabled, and is offered with Zero Data Retention.

Whether the combination of Fable-level performance, lower costs, and tighter controls is enough to hold Anthropic's ground against OpenAI and cheaper Chinese rivals will be worth watching as Sonnet 5.5 and Haiku 5.5 arrive in the weeks ahead.