Claude Opus 5.5 Launches: List Price Cut 20%, Paid Plans’ Five-Hour Limit Raised, 40% Is Anthropic’s Cost Estimate

On September 22, 2026, Anthropic introduced Claude Opus 5.5, the first model in the new “Claude 5.5” family; the announcement page does not mention any region or country restriction. At the same time, Anthropic raised the five-hour usage limits on the Pro, Max, Team, and seat-based Enterprise plans and added one extra rate limit reset; as of the check date, though, claude.com’s pricing page plan comparison table lists the “Opus” series as “No” in the Free column.

This article’s information was checked on September 26, 2026, drawing on Anthropic’s announcement page, the Claude Help Center article on model switching, the claude.com pricing page, and the Opus 5.5 System Card; we have not tested this model ourselves, and this article offers no subscription or upgrade advice.

What happened on September 22: the new family launched on every platform first

Anthropic says Claude Opus 5.5 performs at the level of Claude Fable 5.1 on most work, with an estimated 40% lower cost to run than Opus 5 (conditions in the next section), and it is now available on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure — this passage does not mention subscription plans or countries; on the Claude Platform, developers can get started with the model ID claude-opus-5-5.

On both the API and claude.ai, Opus 5.5 is only available with thinking enabled — it cannot be switched off; and claude.ai, the consumer product, is only available to users aged 18 or above.

Anthropic says this is its first release since it called for “pacing the frontier”; the announcement states that Claude Sonnet 5.5 and Claude Haiku 5.5 “will follow in the coming weeks,” with no date given, and as of the check date they have not yet launched.

Three numbers: what 40%, 20%, and 60% each actually measure

The 40% figure is Anthropic’s own cost estimate, measured “at default settings” on “typical workloads” — it is not a list-price cut; Anthropic explains it comes from Opus 5.5 costing less per token than Opus 5 and using fewer tokens per task, which nets out to a 40% drop in costs.

The 20% and 60% figures are the official list-price cuts: input and output tokens are $4 and $20 per million, 20% less than Opus 5; cache reads are $0.20 per million tokens, 60% less than Opus 5, and Anthropic notes that cache reads make up the majority of agentic and coding work costs.

Anthropic says Opus 5.5 generates output more than 30% faster than Opus 5; Fast mode is also available in Claude Code and the Claude Platform, with up to 2.5x speed, priced at $8 per million input tokens and $40 per million output tokens.

What subscribers should know: the plan comparison table and usage limits

As of the check date, claude.com’s pricing page plan comparison table lists the “Opus” series as “No” in the Free column, and “Yes” across Pro, Max 5x, and Max 20x; Team and both Enterprise plans are also marked “Yes.” The table names the series, not the version number, and the page may change.

The basic rule for usage limits is that every plan has a session limit that resets on a rolling five-hour window, and paid plans add a weekly limit on top; activity on web, desktop, mobile, and Claude Code all draws from the same pool. This time, Anthropic raised the five-hour limit for the Pro, Max, Team, and seat-based Enterprise plans; the announcement doesn’t say by how much or whether it is permanent, and Free is not on the list.

Anthropic is also giving subscription users one rate limit reset, which they can save and use whenever they choose; current usage can be checked anytime under Settings > Usage. Individual plan pricing is $17 per month with the annual subscription discount ($200 billed up front) or $20 per month billed monthly for Pro, and Max starts from $100 per month; these prices don’t include tax and are subject to change.

Why a conversation might suddenly switch to a different model

The Claude Help Center notes that the checks review everything the model reads, not just the user’s latest message — including memory, content from connectors, web search results, and files. Requests flagged as potentially higher-risk offensive cybersecurity work (such as generating exploits or running penetration tests) may fall back from Opus 5 or Opus 5.5 to Opus 4.8, but secure coding work such as scanning source code for vulnerabilities or triaging security issues remains available on Opus 5.5.

Dual-use biology requests in areas like virology, toxicology, and molecular design fall back from Opus 5.5 to Opus 5; everyday health and educational questions (such as interpreting lab results, understanding symptoms, or learning about biology) remain available on Opus 5.5. The frontier LLM development classifiers apply only to Opus 5.5, routing a small number of capabilities related to developing frontier large language models to Opus 5, while general AI/ML development and coding are mostly unaffected. Distillation requests — attempts to extract a model’s internal reasoning — are blocked outright and don’t fall back to another model; asking Claude to explain its reasoning is generally unaffected.

All fallbacks are transparent: you’ll see a notice that the model switched, and the response is labeled with the model that answered. After a switch, the model picker stays on the less capable model for the rest of the conversation — you can switch back yourself from the picker, but if your original request is still part of the conversation, the same safeguards may trigger another fallback. To turn off automatic model switching, go to Settings > Capabilities (Config > MODEL & OUTPUT in Claude Code) and toggle off “Switch models when a message is flagged”; with it off, a flagged request pauses the conversation instead of switching models. The API works differently — automatic switching isn’t active by default, and API customers must opt into and configure the fallbacks themselves.

Vetted organizations can apply starting from the announcement day to the Life Sciences Verification Program to use Opus 5.5 for biology research; the Cyber Verification Program will expand “in the coming weeks,” and as of the check date Opus 5.5 is not yet part of it — most cybersecurity tasks are currently re-routed to Opus 4.8. On billing, requests in the biology, distillation, or frontier LLM development categories are billed when safeguards stop them or hand them to another model before any output; requests blocked before output in other categories are not charged.

How to read the benchmark numbers: who ran them, and what was switched on

Anthropic itself writes in the announcement that, at this level of capability, benchmark margins have become a less reliable guide to real-world differences; in its own use, the gap between Opus 5.5 and Fable 5.1 is narrower than the scores suggest. In the benchmark table, the Terminal-Bench 4.0 figures for GPT-6 Astra and GPT-5.6 Sol are as reported by OpenAI, not run by Anthropic; the AutomationBench score for Opus 5.5 comes from Zapier’s own evaluation during early access.

Opus 5.5 was evaluated with its production safeguards enabled; when they intervened, cybersecurity tasks were completed by Opus 4.8, and biology and frontier LLM development tasks were completed by Opus 5 (except on AutomationBench, which ran without a fallback model, so a safeguard intervention counted as a failure) — Anthropic says this may have lowered Opus 5.5’s benchmark scores. Anthropic also acknowledges seeing signs that Opus 5.5 often suspects it is being evaluated, which makes it harder to judge how the model will behave in real-world settings.

Anthropic says that on prompt injection attacks, Opus 5.5 matches or beats Opus 5 in every setting it tested, including coding, tool use, computer use, and web browsing. The System Card lists one regression, though: the model is more likely than previous models to follow malicious instructions embedded in text that a user pastes into their own prompt. Anthropic says the final snapshot of Opus 5.5, together with some product changes, has mitigated this issue, and it is adding features that remove invisible characters and mark pasted text. Pasting text written by someone else effectively lets that person control part of the prompt.

2026 AI News Roundup: Highlights and Daily Life Applications from January to September

Claude Opus 5 Released: What Changes Matter Most for Everyday Work Users?

• AI tools
• Apps & software
• AI news