Opus 5.5 Is Here, and Your Free Limit Reset Expires Oct 22

On September 22, Anthropic announced its new model on Threads in two sentences: “Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family. It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.” Flagship-level performance on par with Fable 5.1, at 40% less than the previous-generation Opus 5. Then the official Claude account on X added a “One more thing”: five-hour usage limits for subscribers are going up, and everyone gets a “limit reset” they can use whenever they choose.

For paying users, the two things worth remembering from this launch aren’t the benchmarks. They are that Opus 5.5 is now the default model on paid plans, and that the reset expires if you don’t use it by October 22. This piece pulls together the model specs, the pricing, how the reset works and where its limits are, and the compatibility changes API developers need to watch for.

What Opus 5.5 Is: Opus Pricing, Fable Performance

According to Anthropic’s announcement, Opus 5.5 is the first model in the new Claude 5.5 family, pitched as doing Fable 5.1-level work at an Opus-level price. A few of the results Anthropic published:

Benchmark	Opus 5.5	Fable 5.1	Opus 5
Terminal-Bench 4.0	66.4%	55.8%	52.3%
FrontierCode v1.1	54.4%	50.3%	48.0%
GDPval-AA v2.1 (Elo)	1846	1735	1708

On these three, Opus 5.5 doesn’t just match Fable 5.1, it beats it. Anthropic also says it generates output more than 30% faster than Opus 5. Among the early-tester examples cited by 9to5Mac, one team completed a 680,000-line code migration in a single day, and another used it to audit and fix a 200,000-line codebase in three hours.

It is also the first model Anthropic has shipped since CEO Dario Amodei publicly called on the industry to slow the pace of frontier development (we covered the story behind that open letter earlier). The announcement leans on safety numbers: in automated behavioral audits, Opus 5.5 attempted to circumvent boundaries about 85% less often than Opus 5 or Claude Mythos 5.1; most cybersecurity tasks are rerouted to Opus 4.8, and biology requests get the same safeguards as Fable 5.1. In other words, Anthropic chose to compete on “cheaper and faster” rather than “a higher capability ceiling” — consistent with its public stance last week.

Pricing: 20% Cheaper per Token, 60% Off Cache Reads

“40% cheaper” is Anthropic’s figure for overall running cost. If you’re an API customer paying by the token, here’s how the actual unit prices changed (USD per million tokens):

Item	Opus 5.5	Opus 5	Change
Input	$4	$5	−20%
Output	$20	$25	−20%
Cache reads	$0.20	$0.50	−60%
Cache writes (5-minute)	$5	$6.25	−20%

If per-token prices only fell 20%, where does “40%” come from? The key is the 60% cut to cache reads, plus faster output: for workloads that repeatedly reuse the same long context (coding agents, long-document Q&A), the drop in the actual bill will be well beyond 20%. Conversely, if your usage barely hits the cache, expect it to feel roughly 20% cheaper. The context window stays at 1M input tokens and 128K output, the API model ID is claude-opus-5-5, and it’s available on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Azure.

Subscribers: The Same Allowance Stretches 25% Further

For Pro, Max, and Team users, ClaudeDevs put it plainly: “Opus 5.5 is the default for paid plans. It’s priced lower than Opus 5, so your 5-hour and weekly limits go 25% further.”

That 25% is simply the price math: subscription usage limits are metered by cost, so if the unit price drops to 80% of what it was, the same allowance buys 1 ÷ 0.8 = 1.25 times as much work. Anthropic also raised five-hour usage limits for Pro, Max, Team, and seat-based Enterprise plans at the same time, though the announcement didn’t give a specific percentage.

This lands harder when you set it next to earlier changes this month. As we worked out in our previous piece, starting September 14 Claude Code’s weekly limit dropped from 150 during the promotion to a permanent 125 — 83% of the promotional level. Now, running that same 125-unit allowance on Opus 5.5 gets you about 25% more work: 125 × 1.25 ≈ 156, which roughly wins back the effective 17% cut from September 14. That math assumes you stick with the default Opus 5.5; if you manually switch back to Opus 5 or move up to the pricier Fable 5.1, you don’t get the benefit.

How the Reset Works: One Time Only, Expires Oct 22

The most practical piece of this launch is the reset. According to Claude’s Help Center page, “What is a limit reset?”, it’s an occasional offer for eligible plans that restores your usage limit to full immediately when you apply it. To use it:

1. On the web or in Claude Desktop, open Settings → Usage
2. In the Reset section, click the free reset
3. Click again to confirm

The button also appears directly in the message you see when you hit a usage limit.

A few restrictions the Help Center spells out clearly but that are easy to miss:

• One time only, and no undo: once you apply it, it’s gone.
• You don’t have to hit the limit first: but since you only get one, the obvious best time to use it is when you’re genuinely stuck at the cap with urgent work in hand.
• Depending on the reset type shown, it restores either your five-hour limit or your weekly limit; the weekly limit itself still resets on its usual schedule and isn’t pushed back.
• The Claude mobile apps, and Claude Code in the terminal or IDE, don’t have the reset button yet; you can only apply it from a browser or Claude Desktop. Claude Code–only users take note: when you get stuck, you’ll have to switch to the web to press it.
• No refunds, no effect on extra usage: extra usage already billed isn’t refunded, and your usage credit balance doesn’t change.
• If you downgrade or cancel before using it, it’s gone; unused resets expire at the date and time listed in the offer — October 22 this time.

API Developers: Check These Five Things Before Switching

If you call the Claude API from your own code, Opus 5.5 isn’t a painless swap of the model string. According to Anthropic’s developer documentation, migrating from Opus 5 involves several changes that return a straight 400 error:

1. Thinking can’t be turned off: thinking: {type: "disabled"} is rejected at every effort level; the only way to save cost is to lower effort.
2. The effort default changed: it drops from Opus 5’s high to medium. If you were relying on the default, quality may quietly dip after the switch — set it explicitly.
3. No forced tool use: setting tool_choice to any or a specific tool returns a 400; use auto with a prompt instruction, or strict: true / structured outputs.
4. Thinking blocks are bound to the model and the conversation: replaying thinking blocks after editing earlier conversation history may be rejected; design your conversation history to be append-only.
5. The legacy computer-use tool is retired: computer_20251124 returns a 400; you must move to computer_toolset_20260801.

In addition, the notes the model writes between tool calls now come back as thinking blocks rather than text blocks, so if your app only reads text to show progress, it may suddenly go quiet. Anthropic has committed not to retire this model before September 22, 2027.

Bottom Line: Go Find Your Reset Now

To sum up what this means for each kind of user: API customers get a new option that’s 20% cheaper per token and 60% cheaper on cache reads, but have to deal with the compatibility changes first; subscribers don’t need to do anything — the default model has already switched, the same allowance goes about 25% further, and there’s a reset valid until October 22. It’s worth spending ten seconds right now to open Settings → Usage on the web and confirm the reset is already in your account. You only get one, and when you actually need it, you won’t want to be hunting for it.