Anthropic has launched Claude Opus 5.5, the first model in its new Claude 5.5 family. The model matches or outperforms Claude Fable 5.1 across most tasks, but at a 40% lower cost than Opus 5. This is Anthropic’s first model release following the call for “pacing the frontier” by CEO Dario Amodei, and it underwent pre-launch evaluations by external organizations.
Benchmark leadership in coding and knowledge work
According to benchmarks published by Anthropic, Opus 5.5 leads across several key tasks:
-
Terminal-Bench 4.0 (coding agent): Opus 5.5 achieves 66.4%, surpassing Fable 5.1 (55.8%) and GPT-6 Astra (57.9%).
-
CursorBench 4.0 (multi-file coding): 57.8%, beating Fable 5.1 (51.8%).
-
GDPval-AA v2.1 (professional knowledge work): 1,846 Elo, ahead of Fable 5.1 (1,735).
-
Humanity’s Last Exam with tools: 67.7%, exceeding Fable 5.1 (65.6%).
-
AutomationBench (automated workflows): Opus 5.5 records 40.0%, outperforming Fable 5.1 (31.4%) and matching GPT-6 Astra (41.4%).
-
OSWorld 2.0 (computer use): 81.8%, ahead of Fable 5.1 (80.7%).
While these are self-reported figures from Anthropic, external evaluations were conducted prior to launch by Frontier Design and METR.
The most notable highlight lies in cost efficiency. Anthropic stated that Opus 5.5 completed a codebase migration of 680,000 lines in under a day—a task that would take an engineering team several weeks. In internal testing, Opus 5.5 successfully reduced page load times 39 out of 40 times, whereas Opus 5 achieved only minor improvements while altering application behavior.
Sharp price reductions, 30% speed boost
Opus 5.5 is priced at 4 USD per million input tokens and 20 USD per million output tokens, representing reductions of 20% and 25%, respectively, compared to Opus 5. Most notably: cache reads cost just 0.20 USD per million tokens, down 60% from Opus 5. Because cache reads make up the majority of costs in coding and agent workflows, this reduction carries far greater impact than the 20-25% drop on input/output pricing.
Opus 5.5 also generates output 30% faster than Opus 5. A fast mode available in Claude Code runs 2.5 times faster, priced at 8 USD input and 40 USD output.
Anthropic also increased the five-hour usage limits for Pro, Max, Team, and Enterprise plans, and provides subscribers with a rate-limit reset that can be saved and used at any time.
Safety: Highest score on automated behavioral audit
Opus 5.5 achieved the highest score to date on the automated behavioral audit, an alignment testing suite spanning thousands of simulated scenarios. The model is less prone to taking hard-to-reverse actions or acting outside specified boundaries, and resists prompt injection better than Opus 5.
In cybersecurity, Opus 5.5 possesses capabilities on par with Fable 5.1, leading to similar safeguards being applied. Users can identify and fix code bugs within standard software development workflows, but most cybersecurity tasks will be routed to Opus 4.8. The Cyber Verification program will soon expand to include Opus 5.5.
In terms of communication, Opus 5.5 is reported to interact more naturally than Opus 5. It places the most critical information upfront and adheres to user-provided writing rules, making extended study and working sessions easier to follow.
In biology, Opus 5.5 matches or exceeds Claude Mythos 5.1. The Life Sciences Verification program allows verified organizations to use Opus 5.5 for biological research.
Opus 5.5 also debuts with preserved thinking, an anti-distillation safeguard that prevents API users from tampering with Claude’s prior context to extract reasoning traces.
Context: The AI price war unfolds
Opus 5.5 launched just 90 minutes before OpenAI unveiled GPT-6 Sol and Luna. Within 48 hours, Grok 4.7 (SpaceXAI) launched at 2/6 USD, Opus 5.5 at 4/20 USD, and GPT-6 Sol/Luna at 2/10 USD and 0.10/0.50 USD. The coordinated pricing structure that labs maintained through the “pacing the frontier” phase collapsed in less than two days.
Sonnet 5.5 and Haiku 5.5 will follow in the coming weeks, bringing similar improvements in performance, efficiency, and safety. For users, Opus 5.5 offers a compelling new choice: capability matching Fable 5.1 at substantially lower cost, validated by third parties ahead of release.