Claude Sonnet 4.6 in Claude Code: tokens per question
This earlier Sonnet release is a useful baseline for comparing Claude's middle-tier coding models.
The figures
- Bare Claude Code
- 31,400
- With capsul
- 16,600
- less input sent
- x1.89
- answer checks passed out of five, bare then with capsul
- 4 / 4
- tool steps per question, bare then with capsul
- 3.8 / 3
- output tokens per question, bare then with capsul
- 748 / 344
- API-equivalent cost per question, bare then with capsul
- $0.139 / $0.0613
cache-weighted input tokens per question
cache-weighted input tokens per question
90% interval: x1.30 to x2.79
Its API list price is above the newer Sonnet row. If you still run this version for compatibility or because it is the model available in your account, the relevant cost is the combination of its price and the input your agent actually sends. A lower token count on another model does not by itself tell you which one finishes a task better.
The questions below show how much context the bare agent and capsul sent on the same repository tasks. Check the answer results alongside the token interval. They keep a cheaper-looking run from being mistaken for a better run when it failed to find what the task required.
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 | 41,100 | 20,500 | x2.01 | 0 / 0 |
| Q2 | 32,400 | 24,700 | x1.31 | 2 / 3 |
| Q3 | 23,000 | 9,500 | x2.42 | 3 / 3 |
| Q4 | 35,600 | 16,300 | x2.18 | 3 / 2 |
| Q5 | 24,900 | 12,200 | x2.05 | 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 4.6 use per question in Claude Code?
In the September 22, 2026 benchmark, a bare Claude Code session on Claude Sonnet 4.6 sent 31,400 cache-weighted input tokens per one-line coding question on average. With capsul on the same model, 16,600.
How much does a coding question cost on Claude Sonnet 4.6?
At Anthropic list prices, the bare agent's traffic came to $0.139 per question and capsul's to $0.0613. 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 4.6 established?
Yes. The 90% interval runs from x1.30 to x2.79, entirely above one, on 30 measured cells.
Does capsul change Claude Sonnet 4.6's answers?
Not on this protocol: Claude Sonnet 4.6 passed 4 of five answer checks bare and 4 of five with capsul.