Which Claude model uses the least of your limit?

Haiku uses the least of your limit, Sonnet more, Opus more again, and Fable the most. Anthropic's wording is that "Opus costs several times more per turn than Sonnet, and Sonnet more than Haiku", and that Fable models "use them [your limits] faster than other Claude models". Anthropic does not publish the exact ratios, so no one can honestly give you a multiplier.

What Anthropic says

Two Help Center articles carry the ranking:

• On Claude Code usage, Models, usage and limits in Claude Code says Opus costs several times more per turn than Sonnet, and Sonnet more than Haiku. It adds that spending Opus on routine work is the fastest way to drain a daily limit, and that Opus "uses meaningfully more of your quota".
• On Fable, Claude Fable models on your plan says Fable models draw from your plan's regular weekly limits and use them faster than other Claude models.

So the order, cheapest to most expensive against your limit, is Haiku, Sonnet, Opus, Fable. The size of each step is not in any Anthropic source we found.

API prices are a hint, not a limit ratio

Usage credits bill at standard API rates. These are the list prices per million tokens as of November 2026, from claude.com/pricing. Haiku 5.5 is priced by prompt size; the figures shown are for prompts up to 100K tokens, and it costs more above that.

The order matches the ranking above. But these are prices for API billing, not statements about how much of a subscription limit each model consumes. Anthropic has not said that limits scale in the same proportion, so do not turn this table into a prediction.

What you start on

As of November 2026, Opus 5.5 is the default model on Pro, Max, Ultra, Team and Enterprise (since Claude Code v2.1.280; before that, Pro and Team Standard defaulted to Sonnet 5), per the model configuration docs. That means a plain claude on a current version uses the second most expensive tier unless you change it. Fable is never the default; you choose it yourself.

If you are watching your limits, the default is the first thing to look at. How to change it is in how to switch models in Claude Code.

Which model for which job

Anthropic's cost guide gives direct advice: "Sonnet handles most coding tasks well and costs less than Opus. Reserve Opus for complex architectural decisions or multi-step reasoning." For simple subagent tasks it suggests specifying model: haiku.

A practical split that follows:

• Haiku: simple, mechanical subagent work such as searching, summarizing logs or renaming.
• Sonnet: most day-to-day coding.
• Opus: architecture, hard debugging, multi-step reasoning.
• Fable: the hardest and longest-running tasks, when you accept that it uses limits faster.

opusplan: Opus where it counts

opusplan uses Opus in plan mode and Sonnet to execute. Anthropic's Help Center describes this as "plan with Opus, execute with Sonnet". Planning is where a stronger model usually earns its cost, and execution of an agreed plan is where a cheaper one often suffices. Set it with /model opusplan.

One catch from the prompt caching docs: each model has its own cache, so every switch between plan mode and execution under opusplan re-reads the conversation with no cache hits. Plan once, then execute, rather than flipping back and forth in a long conversation.

Haiku for subagents

Subagents still draw on your usage, according to the cost guide, so delegating does not make work free. What it does is let you choose a cheaper model for the delegated part. In a subagent definition you specify model: haiku for simple tasks.

Fable is a different kind of limit

On Max and Ultra, and on premium seats, Fable has its own weekly limit: you can use up to 50% of your weekly limits on Fable models at no extra cost, and that 50% is a share of the same weekly limit, not extra room. On Pro and Team Standard seats, Fable is not included in plan limits and runs on usage credits from the start, as of November 2026. The exception is a Fable 5.2 launch offer that includes Fable models on Pro and Team Standard until November 16, 2026, 11:59 pm PT. See the Fable weekly limit.

What model choice does not fix

Switching models does not restore access once a shared limit is hit. The Claude Code error reference says the session and weekly limits "are shared across all models, so switching models doesn't restore access". It only helps with a limit specific to one model. Effort also matters: higher effort uses more tokens, and you can set it with /effort. Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.2 start at medium effort, per the model configuration docs, so high and above are a deliberate step up in token use.

Some users on GitHub have reported hitting limits quickly on Opus 5.5 even with a single question (issue 105962). Those are reports, not documented behaviour.

Measuring it yourself

Since the ratios are unpublished, the only way to learn what a model costs you is to watch your own bars. Run /usage before and after a comparable task on each model. Treat the result as rough: the bars are account-wide, so anything else using the same account at the same time (claude.ai, another terminal, a script) moves them too, and two tasks are rarely identical.

Headroom, a free independent menu bar app (not made or endorsed by Anthropic), shows each signed-in account's session and weekly percentages in its panel without starting a session, which makes the before and after quick to read. It reads the sign-in Claude Code already saved on your Mac and never sends a prompt. Get Headroom.

For more ways to stretch a limit, read how to make Claude Code limits last longer.