Open Access

Table E.1

Total and stagewise maximum per-query token counts by LLM model and query generation method.

Model Direct Step-by-Step
Total Schema Difficulty Plan SQL Self-Correction Total Schema Difficulty Plan SQL Self-Correction
GPT-5.2 42 671 802 0 0 15363 28877 19 357 802 4 328 8 554 7 991 5 303
GPT-5.3-Codex 39 907 801 0 0 12 261 27 044 22 978 801 4 263 7 039 6 574 6 712
GPT-5.2-Codex 22 106 1 539 0 0 13 740 9 029 35 236 1 539 5 085 12 228 10 697 9 252
GPT-5 26 655 3 919 0 0 13 312 11 232 58 071 3 919 6 609 20 935 16 627 17 157
Claude Opus 4.6 8 193 1 122 0 0 4 716 4 135 13 439 1 122 3 648 5 524 4 415 8 098
GPT-4.1 20 738 795 0 0 8 659 11 351 20 279 951 4 201 7 557 6 878 5 160
Claude Sonnet 3.7 11 303 2 091 0 0 1 211 8 138 8 243 2 091 1 190 1 162 1 153 4 543
Gemini 2.5 Pro 20 741 7 936 0 0 6 826 6 579 34 344 7 936 4 606 8 128 7 499 6 795
GPT-4o 9 509 795 0 0 6 632 4 127 18 677 795 4 809 7 622 4 647 5 010
Claude Sonnet 4.5 5 737 1 729 0 0 947 3 692 12 701 1 729 1 194 2 088 1 101 7 872
Gemini 3 Flash 8 557 1 924 0 0 6 084 3 437 18 789 1 924 4 486 7 458 6 696 6 595
Gemini 3.1 Flash 8 206 913 0 0 7 067 3 397 27 134 913 5 144 7 835 7 982 5 349

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.