Anthropic says Opus 5.5 runs 40% cheaper than Opus 5 as cache reads drop to 20 cents

Anthropic released Claude Opus 5.5 on Tuesday. The company says the model performs on par with its pricier Fable 5.1 on most work, and costs about 40% less than Opus 5 on typical workloads.

Input and output tokens fall to $4 and $20 per million

Anthropic’s launch post says the savings derive from two places. Token prices are down and the model burns fewer tokens per job.

Input tokens are now $4 per million, and output tokens are $20 per million, both 20% cheaper than Opus 5’s $5 and $25. Cache writes declined to $5 per million from $6.25.

The biggest cut was on cache reads, which Anthropic says drive most of the cost in coding and agentic work. They fell 60% to $0.20 per million tokens from $0.50 for Opus 5.

With the base API pricing, Opus 5.5 is about 60% cheaper than Fable 5.1, which sells for $10 and $50 per million tokens. Fable 5.1 had arrived just weeks before.

Anthropic says the model’s output is over 30% faster than Opus 5. Fast mode in the Claude Platform and in Claude Code is up to 2.5x faster, for $8 per million input tokens and $40 per million output.

Opus 5.5 scored 66.4% on Terminal-Bench 4.0, ahead of Fable 5.1 at 55.8%, Opus 5 at 52.3% and 57.9% for OpenAI’s GPT-6 Astra. On GDPval-AA v2.1, a knowledge-work test, Anthropic put it at 1,846 versus Fable’s 1,735.

The company said benchmark margins are a weaker guide to real differences at this capability level.

An early tester did a migration of 680,000 lines of code in less than a day, Anthropic said. Another audited and repaired a 200,000-line codebase in under three hours, work that took Opus 5 more than 20 hours and 2.5 times the tokens.

An internal test showed that Opus 5.5 translated load balancer HAProxy from C to Rust in 9.5 hours. Fable 5.1 took 12 and Opus 5.5 was 51% less costly.

Most cybersecurity tasks reroute to an older Claude model

Anthropic has shipped its first model since CEO Dario Amodei called for pacing frontier AI development on safety grounds, with Opus 5.5. It was examined by external evaluators METR, Frontier Design and others before release.

The model scored the best results of any of its models to date on its automated behavioral audit, said Anthropic. But the company has seen signs that Opus 5.5 often suspects it is being evaluated, which makes it harder to gauge its real-world behavior.

Anthropic says the model is comparable to Claude Mythos 5.1 in biology and cybersecurity, so it comes with safeguards similar to Fable 5.1. Most cybersecurity tasks get rerouted to Opus 4.8, and vetted organizations can apply to a new Life Sciences Verification Program for biology work.

On the same Tuesday, Cryptopolitan said OpenAI has launched GPT-6 Sol and Luna. Sol is $2 per million input tokens and $10 per million output, half of Opus 5.5’s base rates. Luna runs $0.10 and $0.50.

Opus 5.5 registered 40.0% on Zapier’s AutomationBench at a cost of about $1.28 per task, according to Cryptopolitan. The eight-point lead for Anthropic costs about 3.8x as much as GPT-6 Sol at 32.0% for $0.34.

Cryptopolitan has also reported a forecast stating that the costs of inference per agentic workflow will climb by more than five times by 2028.

Opus 5.5 is accessible via the API as claude-opus-5-5 and on Amazon Web Services, Google Cloud and Microsoft Azure.

Anthropic said Claude Sonnet 5.5 and Haiku 5.5 are coming in the next few weeks.