Guide
Which Claude model should you use?
Fable, Opus, Sonnet or Haiku: what each one is for, what it costs, and how to choose without burning through your limits. Based on Anthropic’s and OpenAI’s official documentation.
At a glance
| Model | Best for | Speed | API price | Relative cost |
|---|---|---|---|---|
| Claude Fable 5.1 | The most capable: demanding reasoning, agent sessions that run for hours, research carried through to a finished deliverable. | Slower | $10 / $50 | ×5 |
| Claude Opus 5.5 | The recommended starting point: long-running agentic coding, large refactors, complex systems. | Moderate | $4 / $20 | ×2 |
| Claude Sonnet 5.5 | Everyday work: code generation, data analysis, writing, tool use. | Fast | $2 / $10 | ×1 |
| Claude Haiku 4.5 | The fastest and cheapest: high volumes of simple tasks, sub-agents, real time. | Fastest | $1 / $5 | ×0.5 |
API price in US dollars per million tokens, input / output. Relative cost: price ratio, Sonnet 5.5 = 1. Batch processing costs half, and cache reads a fraction of the input price.
Which model for which task
| Fix a local bug, write a test, rename | Sonnet 5.5, or Haiku 4.5 if it is mechanical |
|---|---|
| Write, summarize, analyze data | Sonnet 5.5 |
| Refactor across files, design an architecture | Opus 5.5 |
| Long autonomous task, research to a finished deliverable | Fable 5.1, when Opus 5.5 falls short even at high effort |
| High volume of small tasks, sub-agents | Haiku 4.5 |
Before switching models, tune the effort
Anthropic puts it plainly: tuning effort is often a better lever than switching models. Opus 5.5 at low effort costs less than Opus at maximum effort, and often does as well as a smaller model. Start with the default effort, and raise it only when a task stalls.
What about your Pro or Max plan?
A Claude plan’s limits are shared between chats and Claude Code: a big coding session reduces what is left for the rest of the day.
Anthropic says it in its help center: “Opus costs several times more per turn than Sonnet, and Sonnet more than Haiku”, without an exact factor. The relative cost above remains the best public order of magnitude.
Some models also have their own ceiling: on Max, Fable models can use up to 50% of your weekly limit; Claude Code applies Opus and Sonnet limits. Switching models does not restore a used-up session: it is shared by all of them.
On Pro and Max, Claude Code uses Opus 5.5 by default. For routine work, /model sonnet spares your limits; the opusplan alias plans with Opus, then executes with Sonnet.
What about ChatGPT with Codex?
| Astra | The most capable, for complex end-to-end work: code, apps, research. |
|---|---|
| GPT-6 Sol | The model for everyday and complex coding and agent work. |
| GPT-6 Luna | The most efficient, for focused, repeatable tasks: summarization, extraction, focused coding. |
OpenAI advises starting from your account’s default setting and raising the reasoning effort when a task needs more planning. Like Anthropic, OpenAI does not publish how much each model weighs on ChatGPT’s limits.
What JaugeAI does for you
- Shows every limit of your plan, including a model’s own weekly limit when there is one, in the Mac menu bar, on iPhone and on Apple Watch.
- Forecasts when you will hit the wall at your pace, and warns you before.
- Advises from your actual usage: for example, trying Opus when long sessions stall on Sonnet.
- Puts a figure on what your activity would cost at API rates, per project.
Free on Mac. On the App Store November 3 for iPhone and Apple Watch.
Frequently asked questions
What is the most powerful Claude model?
Claude Fable 5.1, for demanding reasoning and long agent tasks. Anthropic still recommends starting with Opus 5.5, which suits most work, and moving to Fable 5.1 only when Opus falls short.
Opus or Sonnet for coding?
Sonnet 5.5 for everyday code: fast, and half the API price. Opus 5.5 for refactors, architecture and long autonomous sessions.
Which model uses my limits the least?
Haiku 4.5, then Sonnet 5.5. Anthropic confirms Opus costs several times more per turn than Sonnet, without publishing an exact factor for Pro and Max.
All guides: limits, resets, usage credits