Claude Fable 5 in Claude Code: tokens per question
Fable is Claude's highest-priced tier in this comparison, aimed at work where deeper reasoning may justify the expense.
The figures
- Bare Claude Code
- 41,400
- With capsul
- 17,600
- less input sent
- x2.35
- answer checks passed out of five, bare then with capsul
- 4 / 4
- tool steps per question, bare then with capsul
- 3.8 / 3.2
- output tokens per question, bare then with capsul
- 1,070 / 452
- API-equivalent cost per question, bare then with capsul
- $0.633 / $0.219
cache-weighted input tokens per question
cache-weighted input tokens per question
90% interval: x1.65 to x3.62
Its API list price is above Opus. That changes the decision for ordinary edits: a task that Sonnet can finish cleanly may not need this tier, while a difficult investigation might. Compare the token and answer-check results below with the type of work you actually do before choosing it as a default.
This benchmark row is partial because the subscription limit stopped the run before every planned measurement finished. Its interval has less evidence behind it than a complete row. The missing cells are disclosed in the figures, and the result should be treated as a limited observation rather than a settled saving.
Question by question
Five one-line questions about a real TypeScript codebase, each asked 3 times per arm. Weighted input per question, mean of the repetitions.
| Question | Bare | capsul | Less input | Correct runs, bare / capsul |
|---|---|---|---|---|
| Q1 | 46,100 | 23,500 | x1.96 | 0 / 0 |
| Q2 | 34,900 | 16,600 | x2.10 | 3 / 3 |
| Q3 | 31,700 | 12,600 | x2.51 | 2 / 2 |
| Q4 | 55,300 | 19,000 | x2.91 | 2 / 2 |
| Q5 | 38,800 | 16,100 | x2.41 | 2 / 2 |
List prices
| Model | Input | Output | Cache read | Cache write 5 min | Cache write 1 h | Context |
|---|---|---|---|---|---|---|
| Claude Haiku 4.5 | 1.00 | 5.00 | 0.10 | 1.25 | 2.00 | 200,000 |
| Claude Sonnet 5 | 2.00 | 10.00 | 0.20 | 2.50 | 4.00 | 1,000,000 |
| Claude Sonnet 4.6 | 3.00 | 15.00 | 0.30 | 3.75 | 6.00 | 1,000,000 |
| Claude Opus 4.6 | 5.00 | 25.00 | 0.50 | 6.25 | 10.00 | 1,000,000 |
| Claude Opus 4.7 | 5.00 | 25.00 | 0.50 | 6.25 | 10.00 | 1,000,000 |
| Claude Opus 4.8 | 5.00 | 25.00 | 0.50 | 6.25 | 10.00 | 1,000,000 |
| Claude Opus 5 | 5.00 | 25.00 | 0.50 | 6.25 | 10.00 | 1,000,000 |
| Claude Opus 5.5 | 4.00 | 20.00 | 0.20 | 5.00 | 8.00 | 1,000,000 |
| Claude Fable 5 | 10.00 | 50.00 | 1.00 | 12.50 | 20.00 | 1,000,000 |
| Claude Fable 5.1 | 10.00 | 50.00 | 0.25 | 12.50 | 20.00 | 1,000,000 |
Anthropic list prices, checked September 23, 2026.
How this was measured
Two arms on the same model: Claude Code as anyone runs it, and the same question through capsul. The unit is cache-weighted input: fresh tokens at full weight, cache reads at a tenth. Campaign of September 22, 2026, 3 repetitions per question and arm.
24 / 30 cells · /benchmarks/2026-09-22-claude.json
Questions
How many tokens does Claude Fable 5 use per question in Claude Code?
In the September 22, 2026 benchmark, a bare Claude Code session on Claude Fable 5 sent 41,400 cache-weighted input tokens per one-line coding question on average. With capsul on the same model, 17,600.
How much does a coding question cost on Claude Fable 5?
At Anthropic list prices, the bare agent's traffic came to $0.633 per question and capsul's to $0.219. A Claude subscription does not bill per token: read it as how much of a plan each question uses, compared with other models.
Is the saving on Claude Fable 5 established?
Yes. The 90% interval runs from x1.65 to x3.62, entirely above one, on 24 measured cells.
Does capsul change Claude Fable 5's answers?
Not on this protocol: Claude Fable 5 passed 4 of five answer checks bare and 4 of five with capsul.