Claude Sonnet 5 in Claude Code: tokens per question
Sonnet is Claude's middle tier for coding work that needs more judgement than a narrow lookup.
The figures
- Bare Claude Code
- 54,200
- With capsul
- 18,900
- less input sent
- x2.86
- answer checks passed out of five, bare then with capsul
- 4 / 4
- tool steps per question, bare then with capsul
- 3.2 / 3.2
- output tokens per question, bare then with capsul
- 1,640 / 551
- API-equivalent cost per question, bare then with capsul
- $0.153 / $0.0487
cache-weighted input tokens per question
cache-weighted input tokens per question
90% interval: x2.02 to x4.52
This release carries a lower API list price than the earlier Sonnet row on this site. It is worth considering for routine implementation, tests and review when Haiku needs too much guidance but an Opus model would be an expensive default. The price table below separates fresh input, cached input and output, which need not move together.
The benchmark asks the same short repository questions on both arms. Use its token and answer-check columns to judge this kind of work. A long refactor or an unfamiliar codebase can require a different amount of exploration, so the measured gain is a reference point rather than a promise for every session.
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 | 74,100 | 21,900 | x3.38 | 0 / 0 |
| Q2 | 46,400 | 24,100 | x1.92 | 2 / 3 |
| Q3 | 33,100 | 11,700 | x2.84 | 3 / 3 |
| Q4 | 82,600 | 21,900 | x3.77 | 3 / 3 |
| Q5 | 34,700 | 15,100 | x2.30 | 3 / 3 |
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.
30 / 30 cells · /benchmarks/2026-09-22-claude.json
Questions
How many tokens does Claude Sonnet 5 use per question in Claude Code?
In the September 22, 2026 benchmark, a bare Claude Code session on Claude Sonnet 5 sent 54,200 cache-weighted input tokens per one-line coding question on average. With capsul on the same model, 18,900.
How much does a coding question cost on Claude Sonnet 5?
At Anthropic list prices, the bare agent's traffic came to $0.153 per question and capsul's to $0.0487. 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 Sonnet 5 established?
Yes. The 90% interval runs from x2.02 to x4.52, entirely above one, on 30 measured cells.
Does capsul change Claude Sonnet 5's answers?
Not on this protocol: Claude Sonnet 5 passed 4 of five answer checks bare and 4 of five with capsul.