Claude Opus 4.8 in Claude Code: tokens per question
This later Opus release is a higher-priced option for coding tasks that need a broader investigation.
The figures
- Bare Claude Code
- 65,800
- With capsul
- 13,800
- less input sent
- x4.75
- answer checks passed out of five, bare then with capsul
- 5 / 4
- tool steps per question, bare then with capsul
- 3.4 / 2.2
- output tokens per question, bare then with capsul
- 3,040 / 322
- API-equivalent cost per question, bare then with capsul
- $0.498 / $0.0825
cache-weighted input tokens per question
cache-weighted input tokens per question
90% interval: x3.00 to x8.75
The API list price is the same as the other earlier Opus rows here. When comparing them, the token count and the answer checks are more informative than the family name alone. An agent that opens extra files and revisits earlier tool output can make a seemingly small task expensive even on a fixed model.
These measurements come from brief questions about a real codebase. They show what happened under this protocol, including the variation between questions. For a migration or a long review, measure your own workflow before projecting the displayed saving across a whole 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 | 81,700 | 12,100 | x6.77 | 2 / 0 |
| Q2 | 64,800 | 20,900 | x3.11 | 2 / 3 |
| Q3 | 43,800 | 10,400 | x4.20 | 3 / 3 |
| Q4 | 95,200 | 13,400 | x7.11 | 3 / 3 |
| Q5 | 43,400 | 12,500 | x3.48 | 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 Opus 4.8 use per question in Claude Code?
In the September 22, 2026 benchmark, a bare Claude Code session on Claude Opus 4.8 sent 65,800 cache-weighted input tokens per one-line coding question on average. With capsul on the same model, 13,800.
How much does a coding question cost on Claude Opus 4.8?
At Anthropic list prices, the bare agent's traffic came to $0.498 per question and capsul's to $0.0825. 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 Opus 4.8 established?
Yes. The 90% interval runs from x3.00 to x8.75, entirely above one, on 30 measured cells.
Does capsul change Claude Opus 4.8's answers?
It passed fewer: 5 of five answer checks bare, 4 of five with capsul. The benchmark publishes this rather than leaving it out.