TokenSave
TokenSave Data · Dataset
Claude effort levels: what low, medium, high, xhigh and max cost
The same coding task used 6,299 output tokens and $0.56 at medium effort on Claude Opus 5.5, and 160,033 tokens and $5.25 at max. High cost only 34% more than medium; the jump starts at xhigh.
Measured 2026-10-09 · Opus 5.5 at 5 levels, Sonnet 5.5 at 2 · License CC BY 4.0 · Write-up: Claude Opus 5.5 effort levels tested
DOI 10.5281/zenodo.23446567 · Also on Hugging Face · Kaggle · GitHub
Key findings
- Low and medium are close. Low cost 14% less than medium ($0.48 vs $0.56) and took 32 s instead of 52 s, but built the plainest result.
- High is the best value step. Output tokens rose 96% (6,299 to 12,376) but cost only 34% more, because most of each request is input that does not change.
- Costs jump from xhigh. xhigh cost 143% more than medium and max 837% more. Max used 160,033 output tokens against 6,299 at medium and took 22 min 22 s.
- Sonnet 5.5 is cheaper per run, not just per token. At the same effort it used 35% fewer output tokens than Opus 5.5 and cost 55% less at medium and 60% less at high.
All runs
| Model · effort | Time | Output tokens | Total tokens | Cost | Lines of code |
|---|---|---|---|---|---|
| Opus 5.5 · low | 32 s | 3,368 | 129,315 | $0.48 | 138 |
| Opus 5.5 · medium | 52 s | 6,299 | 133,428 | $0.56 | 358 |
| Opus 5.5 · high | 1 min 50 s | 12,376 | 222,854 | $0.75 | 523 |
| Opus 5.5 · xhigh | 4 min 32 s | 32,435 | 472,082 | $1.36 | 639 |
| Opus 5.5 · max | 22 min 22 s | 160,033 | 2,504,957 | $5.25 | 1,121 |
| Sonnet 5.5 · medium | 37 s | 4,100 | 115,073 | $0.25 | 198 |
| Sonnet 5.5 · high | 58 s | 8,073 | 137,593 | $0.30 | 398 |
Method
- Prompt, word for word: “Make a brick-breaker game that runs in the browser as a single file named index.html in the current folder. It needs 2 levels, a score display and 3 lives, and it must be playable with the keyboard and the mouse. Write the file and finish.”
- Run once per effort level in Claude Code (
claude -p --effort …), each in its own empty folder. - Tokens and cost from ccusage at API prices; time from start and end timestamps; lines counted in the finished
index.html. - All seven games ran without errors and had every requested feature; screenshots and a review of each are in the write-up.
Limitations
- One run per level and one task. Runs vary, especially at high effort levels; treat the numbers as an order of magnitude, not a fixed price.
- Costs are API-equivalent. On a Pro or Max plan the same runs use up your usage limit instead of money.
- Max effort tried to run an automated test that needed permission, which added time.
How to cite
Free to use, quote and republish with attribution under CC BY 4.0. Please link to this page.
TokenSave (2026). Claude effort levels: tokens, time and cost for the same coding task [Data set]. Zenodo. https://doi.org/10.5281/zenodo.23446567
Embed the chart
Related
- Claude Opus 5.5 effort levels tested (full write-up)
- Sonnet 5.5 vs Opus 5.5
- Coding agent cost calculator
Questions or corrections: contact@jonhisking.com