Table F.2
Few-shot PM results by query difficulty when using Claude Opus 4.6 (step-by-step).
| Few-shot | Selection | Claude Opus 4.6 | |||||
|---|---|---|---|---|---|---|---|
| Simple | Medium | Hard | |||||
| Rows | Columns | Rows | Columns | Rows | Columns | ||
| 0-shot | - | 0.97 ± 0.000 | 0.94 ± 0.000 | 0.44 ± 0.070 | 0.72 ± 0.042 | 0.59 ± 0.032 | 0.49 ± 0.057 |
| 1-shot | Random | 0.97 ± 0.000 ~ | 0.93 ± 0.017 ~ | 0.32 ± 0.045 ↓ | 0.72 ± 0.045 ~ | 0.42 ± 0.045 ↓ | 0.46 ± 0.055 ~ |
| QTS | 0.97 ± 0.000 ~ | 0.94 ± 0.000 ~ | 0.30 ± 0.000 ↓ | 0.70 ± 0.000 ~ | 0.48 ± 0.045 ↓ | 0.50 ± 0.000 ~ | |
| MQS | 0.97 ± 0.000 ~ | 0.96 ± 0.017 ~ | 0.42 ± 0.045 ~ | 0.82 ± 0.045 ↑ | 0.52 ± 0.045 ↓ | 0.48 ± 0.045 ~ | |
| 3-shot | Random | 0.90 ± 0.014 ↑ | 0.88 ± 0.014 ↓ | 0.38 ± 0.084 ~ | 0.82 ± 0.045 ↑ | 0.38 ± 0.045 ↓ | 0.42 ± 0.084 ~ |
| QTS | 0.97 ± 0.014 ~ | 0.97 ± 0.000 ↑ | 0.40 ± 0.000 ~ | 0.70 ± 0.000 ~ | 0.48 ± 0.045 ↓ | 0.48 ± 0.045 ~ | |
| MQS | 0.94 ± 0.000 ↑ | 0.94 ± 0.000 ~ | 0.30 ± 0.000 ↓ | 0.70 ± 0.000 ~ | 0.58 ± 0.045 ~ | 0.46 ± 0.055 ~ | |
Notes. The best value per table column is bolded. Arrows compare the few-shot setting against the 0-shot baseline for each query difficulty using two-sided permutation tests on perfect-match scores grouped by run; ↑/↓ indicate significantly higher/lower results at α = 0.05 with no multiple-testing correction, and ~ indicates not significant.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.