Glonce

Anthropic launches Claude Opus 5.5, 40% cheaper than Opus 5

modelsThe Glonce editorsHacker News🔥86Glonce rating86Impact83Novelty68Relevance100Surprise52Players88Credibility92
Anthropic launches Claude Opus 5.5, 40% cheaper than Opus 5

Anthropic introduced Claude Opus 5.5, the first model in its new Claude 5.5 family, calling it the first release since the company said it would pace the frontier. Opus 5.5 performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, and it generates output more than 30% faster.

On performance, Anthropic says early testers saw large jumps on complex work. One tester completed a 680,000-line code migration in less than a day, a job the company says would have taken an engineering team weeks. Asked to cut load times across every page of a web app, Opus 5.5 succeeded 39 of 40 times, while Opus 5 made smaller improvements that also changed the app's behavior. In a separate test, several Claude models built a game from a single prompt, and Opus 5.5 scored highest on graphics and polish.

On safety, Anthropic says Opus 5.5 achieves the best scores of any model it has tested on its automated behavioral audit, an alignment suite covering thousands of simulated scenarios, and that it is less likely than recent models to take hard-to-reverse actions or act outside the boundaries it is given. It is also more resistant than Opus 5 to prompt injection, in every setting Anthropic tested, including coding, tool use, computer use and web browsing; on a benchmark run by the AI security firm Gray Swan, it ties Claude Fable 5.1 for the lowest prompt injection success rate of any model tested. The model was tested before release by external evaluators including Frontier Design and METR. Because Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, Anthropic is deploying it with safeguards similar to those on Claude Fable 5.1: vetted organizations can apply now to the Life Sciences Verification Program to use it for biology research, and verified cybersecurity practitioners will get access through the Cyber Verification Program in the coming weeks.

On cost, standard pricing is $4 per million input tokens and $20 per million output tokens, 20% less than Opus 5, and cache reads, which Anthropic says make up most agentic and coding costs, are $0.20 per million tokens, 60% less than Opus 5. A Fast mode in Claude Code and the Claude Platform runs at up to 2.5x speed for $8 per million input tokens and $40 per million output tokens. Anthropic is also raising five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans and giving subscribers a rate-limit reset they can save and use whenever they choose.

On coding, an early tester used Opus 5.5 to audit and fix a 200,000-line codebase in under three hours, versus over 20 hours and 2.5 times the tokens for Opus 5. In an internal test translating HAProxy from C to Rust, both Opus 5.5 and Claude Fable 5.1 passed nearly all of HAProxy's own regression tests, but Opus 5.5 finished in 9.5 hours against 12 for Fable 5.1, at 51% less cost. At default effort, Opus 5.5 beats GPT-6 Astra on FrontierCode for roughly 20% of the cost per task, matches Astra on Terminal Bench 4.0 for about 40% of the cost, and beats GPT-5.6 Sol on CursorBench by 11 points for about a third of the cost. Enterprise testers echoed the efficiency claims: GitHub's Mario Rodriguez said Opus 5.5 used among the fewest tokens and steps the company measured, solving more terminal tasks in VS Code than Opus 5 in less than half the steps; Clio's Sean Heintz said it ran unattended for over 18 hours on a six-repository task with minimal reworking; Quantium's Harley Barnes said a task that previously took 38 prompts over four days came in at 11 prompts over three hours; Optiver's Noyan Tokgozoglu said it cut the cost of an agentic coding workload by 40 to 50% while matching Opus 5's quality in about half the turns, time and tokens; and Kiro's Deepak Singh said it solved more of a public command-line benchmark than Opus 5 while making about 40% fewer calls and using half the tokens.

On knowledge work, Anthropic asked Opus 5.5, Claude Fable 5.1 and Opus 5 to write a report on a company's quarterly performance from a version of the web where the earnings release was hard to find; across different effort settings, 16 of 18 of Opus 5.5's reports cleared Anthropic's quality bar, where any invented figure or quote would have failed, and neither Fable 5.1 nor Opus 5 cleared it on any attempt. Investment firm Walleye Capital, an early tester, said Opus 5.5 largely solved its evaluation suite on the lowest setting and, on higher settings, caught and corrected an error in the evaluation instructions that no other model had found. Asked to analyze a fictional merger between two HR software companies, both Opus 5.5 and Opus 5 reached the same conclusion, but Opus 5.5's financial model and presentation were more thorough and easier to read, and it finished in 63 minutes against 93 for Opus 5, at 50% less cost. On GDPval-AA v2.1, a benchmark of real-world work across 44 occupations, Opus 5.5 scores 1846 Elo, ahead of Fable 5.1 and Opus 5, and at its default, medium effort it beats GPT-6 Astra running at max effort for about a fifth of the cost per task.

On security for coding agents, Anthropic says Opus 5.5 ships with a classifier that screens every action before it runs, an open source sandbox that security teams can audit, and code review meant to catch vulnerabilities before they merge.

Claude Sonnet 5.5 and Claude Haiku 5.5 are due to follow in the coming weeks with many of the same improvements to performance, efficiency and safety. Anthropic gives no specific release date for either model, nor a general availability date for anything beyond Opus 5.5 itself.

Key facts

  • Opus 5.5 costs 40% less to run than Opus 5 at default settings; standard pricing is $4/$20 per million input/output tokens, and cache reads are $0.20 per million tokens, 60% less than Opus 5.
  • One tester completed a 680,000-line code migration in under a day, and another audited and fixed a 200,000-line codebase in under three hours, versus over 20 hours for Opus 5.
  • On GDPval-AA v2.1, a benchmark across 44 occupations, Opus 5.5 scores 1846 Elo, and it beats GPT-6 Astra on FrontierCode for roughly 20% of the cost per task.
  • External evaluators Frontier Design and METR tested Opus 5.5 before release; Anthropic says it posts the best scores yet on its automated behavioral audit and matches or beats Opus 5 against prompt injection in every setting tested.
  • Claude Sonnet 5.5 and Claude Haiku 5.5 are due in the coming weeks; Pro, Max, Team and seat-based Enterprise plans get higher five-hour usage limits and a savable rate-limit reset.

Why it matters

Opus 5.5 is Anthropic's first release since it said it would pace the frontier, and the company frames it as an efficiency jump rather than a pure capability leap: it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5 and generating output more than 30% faster. The headline examples are about doing large jobs faster and cheaper, such as a 680,000-line code migration finished in under a day, and fixing every page's load time across a web app with 39 of 40 successful attempts, against smaller and behavior-altering fixes from Opus 5.

Who it affects

Developers and engineering teams working through Claude Code, GitHub Copilot CLI, VS Code and the Claude Platform, including the early testers Anthropic names: GitHub, Clio, Lovable, Quantium, Spotify, Optiver, Column and Kiro. Pro, Max, Team and seat-based Enterprise subscribers get higher five-hour usage limits and a rate-limit reset they can save for later. Vetted biology researchers can apply now for the Life Sciences Verification Program, and verified cybersecurity practitioners get Cyber Verification Program access in the coming weeks. Teams not moving to Opus can expect Claude Sonnet 5.5 and Claude Haiku 5.5 with similar improvements in the coming weeks.

How to use it

Standard pricing is $4 per million input tokens and $20 per million output tokens, 20% less than Opus 5, with cache reads, which Anthropic says cover most agentic and coding costs, at $0.20 per million tokens, 60% less than Opus 5. A Fast mode in Claude Code and the Claude Platform runs at up to 2.5x speed for $8 per million input tokens and $40 per million output tokens. Verified organizations can apply now for the Life Sciences Verification Program, and cybersecurity practitioners get Cyber Verification Program access in the coming weeks.

How solid is it

The evidence mixes Anthropic's own benchmarks, an internal test suite and customer-supplied testimonials. Benchmarks cited include GDPval-AA v2.1 (1846 Elo across 44 occupations), FrontierCode, Terminal Bench 4.0 and CursorBench against GPT-6 Astra and GPT-5.6 Sol, plus a HAProxy C-to-Rust rewrite finished in 9.5 hours against 12 for Fable 5.1. External evaluators Frontier Design and METR tested the model before release, and the AI security firm Gray Swan ran the prompt injection benchmark cited. The eight enterprise quotes (GitHub, Clio, Lovable, Quantium, Spotify, Optiver, Column, Kiro) are testimonials Anthropic selected and published, not an independent study, and Anthropic itself says that at this capability level, benchmark margins have become a less reliable guide to real-world differences; in its own use, it says the gap to Claude Fable 5.1 is narrower than the scores suggest.

Risks and caveats

Because Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, Anthropic is gating some of that capability behind verification programs rather than making it open by default. No general availability date is given for Opus 5.5 itself, and Sonnet 5.5 and Haiku 5.5 are promised only as 'in the coming weeks,' with no pricing yet disclosed for either. The full System Card is referenced but not reproduced in the source, and Anthropic acknowledges its expanded alignment testing, covering longer tasks, impossible tasks and scenarios modeled on real incidents, 'still has limits.'

“Developers want agents that can take on real software work and finish it. In our testing across GitHub Copilot CLI and VS Code, Claude Opus 5.5 used among the fewest tokens and steps we measured.”

— Mario Rodriguez, Chief Product Officer, GitHub