September 25, 2026. Anthropic released Claude Opus 5.5 on September 22, the first model in a new Claude 5.5 family, and several headlines the same day said it had scrapped the five-hour usage caps on Claude subscriptions. Anthropic's own announcement says something narrower: it is increasing the 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 when they choose. For an agency running Claude Code across a team, that distinction is the difference between planning around a bigger window and assuming there is no window at all. Here is what the release actually says, read from Anthropic's launch page and pricing documentation on September 25.
What changed
- Price. Input is $4 per million tokens and output $20, each 20% below Opus 5's $5 and $25. Cache reads are $0.20 per million, 60% below Opus 5's $0.50, and Anthropic's pricing page lists that as a 0.05x multiplier on Opus 5.5 against the usual 0.1x. Cache writes are $5 for the five-minute cache and $8 for the one-hour cache. Anthropic says the combination costs 40% less than Opus 5 on typical workloads at default settings, because cache reads make up the majority of agentic and coding spend.
- Fast mode. $8 input and $40 output for up to 2.5x speed, in Claude Code and the Claude Platform, first-party API only.
- Usage limits. Five-hour limits increased on Pro, Max, Team and seat-based Enterprise, plus a saveable rate limit reset for subscription users. The limits still exist.
- Context and output. 1M token context window and 128K output, with the full window billed at standard pricing; Anthropic's pricing page says a 900K token request costs the same per token as a 9K one.
- Speed and efficiency. Output generation more than 30% faster than Opus 5, and fewer tokens per task. Anthropic's testers report a 200,000-line codebase audited and fixed in under three hours where Opus 5 took over 20 hours and 2.5x the tokens.
- Safeguards. Because Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, it ships with safeguards similar to Claude Fable 5.1; cybersecurity practitioners get access through the Cyber Verification Program in the coming weeks. Anthropic says it is the strongest model to date on its automated behavioral audit and more resistant than Opus 5 to prompt injection.
- What is next. Sonnet 5.5 and Haiku 5.5 follow in the coming weeks. Opus 5.5 is committed to be available until at least September 22, 2027.
How the 40% actually shows up
The headline rates fell 20%, so the other half of the saving comes from two places. First, the cache read rate: an agent whose turns are mostly a cached system prompt and conversation history now pays $0.20 instead of $0.50 per million for the bulk of its input. Second, token efficiency: Anthropic's numbers say Opus 5.5 finishes the same tasks in fewer steps and fewer output tokens, so a task that used to cost a dollar of output now costs less even before the rate cut. Both savings depend on how the workflow is built. A prompt that changes its prefix on every call gets none of the cache benefit, which is why the cache multiplier matters more than the list price for any agent that runs all day.
What it means for operators
If your team lives in Claude Code, the practical changes are a larger five-hour window, a reset you can bank for the day a deadline lands, and a Fast mode that costs double when a client is waiting on the call. If you build client automations on the API, the move is to migrate long-running agents first, because they carry the most cached tokens and gain the most from the $0.20 read rate, and to leave short one-shot prompts where they are until Sonnet 5.5 arrives. For agencies that resell AI features inside their own product, the arithmetic changes what you can bundle: a per-seat AI assistant that cost forty cents of model time a day now costs closer to twenty-five, which is the kind of margin shift that decides whether a feature ships in a white-label SaaS offer or stays on the roadmap. Whichever side you are on, read the vendor page rather than the headline: the caps were raised, not removed, and the price is 40% lower only for workloads shaped like the ones Anthropic tested.
Frequently Asked Questions
$4 per million input tokens and $20 per million output, with cache reads at $0.20 per million, cache writes at $5 (five-minute) or $8 (one-hour), and Fast mode at $8 and $40. Anthropic says typical workloads cost 40% less than on Opus 5 at default settings.
No. Anthropic's announcement says it is increasing five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans, and giving subscription users a rate limit reset they can save and use when they choose. The limits still exist; they are larger.
No. Anthropic's pricing page says the full 1M token context window is billed at standard pricing, so a 900K token request costs the same per token as a 9K one. Prompt caching and batch discounts also apply across the full window.
Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks with many of the same performance, efficiency and safety improvements. No date has been published.