Anthropic launches Claude Opus 5.5 with 40% price cut

Anthropic released Claude Opus 5.5 on September 22, 2026, introducing significant price reductions and performance improvements to its AI model lineup. The new model costs 40% less to run than its predecessor Opus 5, with pricing set at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 respectively. Claude Code users received a 20% increase in their five-hour session limits, and the lower pricing means those limits effectively extend about 25% further.

Price Cuts and Performance Gains

Anthropic says Opus 5.5 performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5. Cache reads, a critical cost factor for coding agents, dropped 60% from $0.50 to $0.20 per million tokens.

In early testing, Opus 5.5 resolved tasks comparably to Claude Opus 5 while using significantly fewer steps and tokens, and it quickly recovered from errors in multistep tasks. The model tops Anthropic’s benchmark table on SWE-bench Pro at 89.9% and Terminal-Bench 4.0 at 66.4%.

Claude Code Gets Default Model Upgrade

Claude Opus 5.5 became the default model in Claude Code immediately upon launch. Pro, Max, and Team plan subscribers received a one-off rate-limit reset that they can save and use whenever they choose.

The model maintains a 1-million-token context window, 128K max output tokens, and a knowledge cutoff of June 2026, with adaptive thinking always on and a default effort of medium. The Batch API supports up to 300K output tokens with a beta header.

Platform Availability and Technical Changes

Claude Opus 5.5 is now available on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. The model also launched in GitHub Copilot on September 22.

Anthropic says Opus 5.5 puts important information first, uses less jargon and follows writing rules better, addressing complaints about earlier models. The model debuts with a new output style called “Clear, conversational English, with no Claude-isms,” which eschews eager openers, emojis, corporate vocabulary, and wrap-ups.

However, developers need to be aware of breaking changes. Progress text that previously streamed between tool calls now arrives as thinking blocks that are empty at the default display setting, marked as “omitted”.

New Safety Measures

Opus 5.5 launches with preserved thinking, an anti-distillation safeguard introduced with Fable 5.1. This feature stops API users from editing Claude’s prior context to extract reasoning, and applies to Opus 5.5 for API accounts created on or after August 31, 2026.

Claude Opus 5.5 watermarks its text outputs, though the watermark doesn’t change the meaning, quality, or readability of outputs, nor does it add any tokens or cost.

Organizations conducting research and development work impeded by biology safeguards can apply to Anthropic’s new Life Sciences Verification Program, which gives vetted organizations like academic labs, startups, and pharmaceutical companies access to safeguards designed for the full breadth of biology-related work.

Token Usage Trade-offs

While the model offers substantial cost savings, it comes with increased token consumption. The model requires 119,000 tokens on average to complete a given Artificial Analysis Intelligence Index task, compared to just 17,000 tokens for OpenAI’s GPT-6 Astra (xhigh) for the same typical task.