Claude hands paid users a spare limit reset until Oct 22

Anthropic has given Claude Code a spare tank. Paid users get one usage-limit reset that they can save and spend any time before Oct 22, the ClaudeDevs account said on Twitter. It comes with Opus 5.5, a new default model that Anthropic says performs at the level of Fable 5.1 and costs less than Opus 5.

At a glance

• Opus 5.5 is now the default on paid plans, and Anthropic says it performs at Fable 5.1's level while running about 30% faster and about 40% cheaper per task than Opus 5.
• Five-hour session limits in Claude Code rose 20% today, and separately, the lower Opus 5.5 price makes both the 5-hour and weekly limits go 25% further.
• Opus 5.5 carries Fable 5.1-class safeguards on cyber, bio and frontier LLM development. Flagged requests fall back to another model, and Anthropic concedes that some of those flags are wrong.

A quick refresher on how the plans work: Claude Code on paid plans runs against a 5-hour session limit and a weekly limit, and once either is used up, you wait for it to roll over. This release is measured against two other models. Anthropic compares Opus 5.5 with its predecessor, Opus 5, on price and speed, and with Fable 5.1 on capability.

A saved reset you can use any time until Oct 22

If you are on Pro, Max or Team, the reset is already in Settings → Usage. You can use it the moment a session runs dry or keep it for a harder week, as long as you apply it before Oct 22.

Testing Catalog reported that the same Reset button has also appeared on Claude web and Desktop, where it invites users to explore Opus 5.5. The outlet added that, in its view, the button looks like a testing button.

The reset is not the only extra capacity. Five-hour session limits in Claude Code went up 20% today. Anthropic lists that increase separately from the gain that comes with the new default model, which we cover below.

Opus 5.5 costs $4/$20 per million tokens on the Claude Platform, down from $5/$25

On the Claude Platform, Opus 5.5 costs $4/$20 per million tokens, compared with $5/$25 for Opus 5. Cache reads drop further, by 60%, from $0.50 to $0.20. That matters most for agents that send the same long context again on every turn.

If you care more about speed than savings, fast mode goes up to 2.5x and costs $8/$40. Anthropic also says Opus 5.5 is about 30% faster and about 40% cheaper per task than Opus 5, and that it performs at the level of Fable 5.1.

Switching takes one command: run /claude-api migrate in Claude Code to move to claude-opus-5-5. Anthropic also points to a post on its blog about cutting costs on the Claude Platform. The model is available now in Claude Code and on the Claude Platform.

Thinking is always on for Opus 5.5, with medium effort as the default

Opus 5.5 always thinks before it answers, and you cannot turn that off, only tune it. Medium is the default level, and the effort parameter lets you raise or lower it depending on how hard the task is.

For long-running agents, Anthropic points to compaction, which summarizes older turns of a conversation and keeps the recent ones intact. The Claude Platform documentation explains how to set it up for agents that run long enough to fill their context.

Anthropic also says the model writes better. According to the post, it is clearer, uses less jargon and sticks more closely to the writing rules you give it, so long Claude Code sessions are much easier to follow.

Why a lower price makes the same limit go 25% further

The ClaudeDevs post links the 25% gain directly to price. Opus 5.5 costs less than Opus 5, so the same 5-hour and weekly limits cover more work. The numbers match the Platform prices: at four-fifths of the old per-token price, a fixed budget buys a quarter more tokens.

Picture a fixed grocery budget when the shop cuts its prices by a fifth. The budget stays the same, but you take home more food. The 20% session increase makes the budget itself bigger, and the saved reset works like a second budget that you decide when to spend.

The safeguards work by sending some requests elsewhere. Opus 5.5 is the first Opus with Fable 5.1-class safeguards on cyber, bio and frontier LLM development. When a request is flagged, it falls back to another model and Opus 5.5 does not answer it. Anthropic describes safeguards as one way it prevents misuse as its models become more powerful.

The post leaves some gaps. It does not say which model handles flagged requests, how often the flags are wrong, or how a 40% per-task saving fits with a one-fifth per-token cut, which likely means Opus 5.5 uses fewer tokens per task. In our view, the fallback is the part to watch: a wrong flag on legitimate security work would quietly move your request to a different model.

Before the reset expires on Oct 22

Your reset can be used until Oct 22. The post does not say whether another one will follow or whether saved resets will become a regular feature. Anthropic says it is working to reduce incorrect safeguard flags but gives no timeline or target for that work. Testing Catalog's view that the web and Desktop button looks like a test also leaves open whether the button will stay after the deadline.