Claude Opus 5 in Claude Code: tokens per question
This Opus release is a higher-priced Claude option for coding work that benefits from a longer chain of decisions.
The figures
- Bare Claude Code
- 45,300
- With capsul
- 16,200
- less input sent
- x2.80
- answer checks passed out of five, bare then with capsul
- 4 / 4
- tool steps per question, bare then with capsul
- 4.8 / 2.8
- output tokens per question, bare then with capsul
- 1,370 / 520
- API-equivalent cost per question, bare then with capsul
- $0.343 / $0.101
cache-weighted input tokens per question
cache-weighted input tokens per question
90% interval: x1.95 to x4.18
Its API list price matches the earlier Opus releases and sits above the newer Opus row. It may be justified for ambiguous bugs, architecture work or cross-file changes, but the agent's exploration still determines much of the input spent. The page separates what the bare agent sent from what it sent with capsul so that the comparison remains inspectable.
Treat this campaign's gain with care. A separate complete run on this model did not reproduce the same result. The published row remains part of the shared campaign for comparability, while the interval and task rows show what this particular run supports.
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 | 58,000 | 22,600 | x2.57 | 0 / 0 |
| Q2 | 42,800 | 18,200 | x2.36 | 2 / 3 |
| Q3 | 35,100 | 10,500 | x3.34 | 3 / 3 |
| Q4 | 47,100 | 16,700 | x2.82 | 3 / 3 |
| Q5 | 43,500 | 12,900 | x3.36 | 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 5 use per question in Claude Code?
In the September 22, 2026 benchmark, a bare Claude Code session on Claude Opus 5 sent 45,300 cache-weighted input tokens per one-line coding question on average. With capsul on the same model, 16,200.
How much does a coding question cost on Claude Opus 5?
At Anthropic list prices, the bare agent's traffic came to $0.343 per question and capsul's to $0.101. 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 5 established?
Yes. The 90% interval runs from x1.95 to x4.18, entirely above one, on 30 measured cells.
Does capsul change Claude Opus 5's answers?
Not on this protocol: Claude Opus 5 passed 4 of five answer checks bare and 4 of five with capsul.