Claude Opus 5.5 ships with Fable-level performance at 40% lower cost, lands on Vercel AI Gateway

Anthropic has released Claude Opus 5.5, the first model in its new Claude 5.5 family, claiming it performs at the level of its larger Fable 5.1 model on most work while costing roughly 40% less per task than Opus 5. The model is available immediately, and Vercel has added it to its AI Gateway on launch day.

Cheaper and faster

According to Anthropic's announcement, input and output tokens are priced at $4 and $20 per million respectively, 20% below Opus 5, while cache reads fall 60% to $0.20 per million — notable because cached tokens make up the majority of costs in agentic and coding workloads. The model also generates output more than 30% faster than its predecessor, and a fast mode offering up to 2.5x speed is available in Claude Code and on the Claude platform at $8 and $40 per million input and output tokens.

Anthropic is additionally raising five-hour usage limits on Pro, Max and Team plans, and giving subscribers a rate-limit reset they can hold and use whenever they choose. TechCrunch points out that the launch lands just two months after Opus 5 arrived on July 24.

Performance and efficiency claims

Anthropic calls Opus 5.5 its new leading model, with the biggest gains in agentic coding, long-running agent tasks and knowledge work. In the company's own benchmark tables it scores 66.4% on Terminal-Bench 4.0 and 54.4% on FrontierCode v1.1 — ahead of Fable 5.1, Opus 5 and OpenAI's GPT-6 Astra as reported by Anthropic — plus 81.8% on the OSWorld 2.0 computer-use suite. Anthropic itself cautions that at this capability level benchmark margins are a less reliable guide to real-world differences, and says that in its own use the gap to Fable 5.1 is narrower than the scores suggest. TechCrunch, meanwhile, reports that Opus 5.5 outpaces Fable on many benchmarks and succeeded at informal tasks Fable failed to complete.

Early testers cited by Anthropic include one who completed a 680,000-line code migration in under a day, and another who audited and fixed a 200,000-line codebase in under three hours — work that took Opus 5 more than 20 hours and 2.5x the tokens. In an internal test translating the HAProxy load balancer from C to Rust, Opus 5.5 finished in 9.5 hours against Fable 5.1's 12, at 51% lower cost. Anthropic also says the model writes more plainly and puts key information up front, which it frames as both a usability and a safety benefit because output is easier to verify.

Safety and the pacing context

TechCrunch notes this is Anthropic's first release since CEO Dario Amodei endorsed deliberately slowing capability advances so risk-prevention work can keep up. Anthropic says Opus 5.5 was evaluated before release by outside organizations including METR and Frontier Design, and scored higher than any model it has tested to date on its automated behavioral audit, an alignment suite spanning thousands of simulated scenarios. The company describes it as less likely than recent models to take hard-to-reverse actions and more resistant to prompt injection than Opus 5, while acknowledging it still has limits.

Because Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, it ships with the same safeguards as Fable 5.1. Vetted organizations can already apply to the Life Sciences Verification Program, and Anthropic says the Cyber Verification Program will open to verified security practitioners in the coming weeks. TechCrunch reports the underlying safety training was broadly similar to previous models, with stronger evaluation and monitoring systems being prepared for future releases.

On Vercel AI Gateway, with breaking changes

Vercel announced same-day availability on AI Gateway under the model ID anthropic/claude-opus-5.5, with fast mode exposed as anthropic/claude-opus-5.5-fast. The model is reachable through the AI SDK, the OpenAI-compatible Chat Completions API, the Anthropic Messages API and connected coding agents, with a 1M-token context window, up to 128K output tokens and a June 2026 knowledge cutoff. Vercel passes provider pricing through with no markup or platform fee, and offers opt-in regional inference — pinning requests to the US, for example — and Zero Data Retention.

Vercel warns of two API changes that can turn previously valid requests into HTTP 400 errors: thinking is now always adaptive, so requests that disable thinking or set a fixed thinking budget are rejected, and forced tool use is retired. Developers are told to steer the model with prompts and effort settings, retry missed tool calls in their own harness, and switch to structured outputs where forced tool calls were previously used to produce JSON.

Anthropic says Sonnet 5.5 and Haiku 5.5 will follow in the coming weeks with similar improvements to performance, efficiency and safety.

Why it matters

Agentic coding bills scale with token volume, and Opus 5.5 cuts cost on two axes at once — lower per-token prices and fewer tokens per task — so the claimed 40% net saving matters more than any single benchmark point. The release is also the first concrete test of Amodei's stated commitment to pacing the frontier, pairing external evaluation with capability-matched safeguards rather than capability gains alone. For engineering teams, the retired forced-tool-use and fixed-thinking-budget options mean real migration work, and day-one availability through Vercel's gateway at no markup underscores how quickly frontier models now reach third-party distribution.