Ranking run 2026-05-10
Claude Opus 4.7 leads with 76/100 across all domains. Tax law: Claude Opus 4.7 (71) · Medicine: Claude Opus 4.7 (83) · Law: Claude Opus 4.7 (77) · Business law: GPT-5 (79).
Overall across domains
| # | Model | top100 | 95 % CI | Δ prev. | failures | sec median/p90 | hallu band | per 1000 tok. | Δ own family |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | 0 | 17 / 31 | 76 | 71–80 | -1 | 53 | 1,48 | — |
| 2 | GPT-5 | 0 | 29 / 38 | 73 | 67–78 | +6 | 65 | 0,62 | — |
| 3 | Mistral Large 2 | 0 | 33 / 369 | 45 | 36–54 | -1 | 39 | 4,43 | — |
| 4 | Gemini 2.5 Pro | 0 | 20 / 24 | 44 | 38–50 | +23 | 72 | 4,14 | — |
Δ prev. = difference to the previous published run; “—” = model was not in it. Hallu band: 100 = none, 0 = ≥ 4 false claims per answer. Per 1000 tok. = flagged claims per 1000 output tokens (length-normalised). Δ own family = score with all judges minus score with foreign judges only, in points; positive = own family scores more generously; “—” = not recorded before code_version 0.2.8 or no foreign judge (methodology §4, §6.5).
Tax law · 8 Items
Ranks 1–2 statistically tied (95 % CI overlaps rank 1)
In Tax law, Claude Opus 4.7 (top score) contradicts a stored key fact on average 0,6 times per answer and adds avg 1,7 claims the judge panel flags as false or unsupported; 100 % of answers carry at least one flag. The lowest value is Gemini 2.5 Pro at avg 1,8 per answer. Relative to answer length: 1,70 per 1000 output tokens, lowest GPT-5 at 0,89.
| # | Model | top100 | 95 % CI | Δ prev. | contradicts key fact | extra false/unsupp. | avg total | answers affected | per 1000 tok. | IRR |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | 71 | 61–80 | +3 | 0,6 | 1,7 | 2,3 | 100 % | 1,70 | 80 % |
| 2 | GPT-5 | 55 | 44–68 | +13 | 1,4 | 1,1 | 2,5 | 100 % | 0,89 | 75 % |
| 3 | Gemini 2.5 Pro | 32 | 19–46 | +32 | 1,3 | 0,8 | 1,8 | 100 % | 7,73 | 61 % |
| 4 | Mistral Large 2 | 9 | 4–13 | +2 | 2,9 | 1,3 | 4,2 | 100 % | 9,87 | 58 % |
Medicine · 9 Items
Ranks 1–3 statistically tied (95 % CI overlaps rank 1)
In Medicine, Claude Opus 4.7 (top score) contradicts a stored key fact on average 0,3 times per answer and adds avg 0,6 claims the judge panel flags as false or unsupported; 100 % of answers carry at least one flag. The lowest value is Gemini 2.5 Pro at avg 0,5 per answer. Relative to answer length: 0,89 per 1000 output tokens, lowest GPT-5 at 0,60.
| # | Model | top100 | 95 % CI | Δ prev. | contradicts key fact | extra false/unsupp. | avg total | answers affected | per 1000 tok. | IRR |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | 83 | 73–90 | +1 | 0,3 | 0,6 | 0,9 | 100 % | 0,89 | 79 % |
| 2 | GPT-5 | 82 | 75–88 | -1 | 0,2 | 0,7 | 0,9 | 89 % | 0,60 | 77 % |
| 3 | Mistral Large 2 | 64 | 52–76 | -4 | 0,3 | 1,0 | 1,3 | 100 % | 2,23 | 74 % |
| 4 | Gemini 2.5 Pro | 54 | 47–61 | +25 | 0,3 | 0,3 | 0,5 | 67 % | 2,16 | 69 % |
Law · 8 Items
Ranks 1–2 statistically tied (95 % CI overlaps rank 1)
In Law, Claude Opus 4.7 (top score) contradicts a stored key fact on average 0,3 times per answer and adds avg 0,9 claims the judge panel flags as false or unsupported; 100 % of answers carry at least one flag. The lowest value is Gemini 2.5 Pro at avg 0,4 per answer. Relative to answer length: 1,41 per 1000 output tokens, lowest GPT-5 at 0,26.
| # | Model | top100 | 95 % CI | Δ prev. | contradicts key fact | extra false/unsupp. | avg total | answers affected | per 1000 tok. | IRR |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | 77 | 69–85 | -5 | 0,3 | 0,9 | 1,2 | 100 % | 1,41 | 83 % |
| 2 | GPT-5 | 73 | 66–80 | +9 | 0,0 | 0,5 | 0,5 | 75 % | 0,26 | 68 % |
| 3 | Mistral Large 2 | 54 | 47–63 | +1 | 0,4 | 0,6 | 1,0 | 88 % | 3,01 | 62 % |
| 4 | Gemini 2.5 Pro | 51 | 43–61 | +34 | 0,3 | 0,2 | 0,4 | 50 % | 1,95 | 75 % |
Business law · 8 Items
Ranks 1–2 statistically tied (95 % CI overlaps rank 1)
In Business law, GPT-5 (top score) contradicts a stored key fact on average 0,4 times per answer and adds avg 1,3 claims the judge panel flags as false or unsupported; 100 % of answers carry at least one flag. Relative to answer length: 0,60 per 1000 output tokens.
| # | Model | top100 | 95 % CI | Δ prev. | contradicts key fact | extra false/unsupp. | avg total | answers affected | per 1000 tok. | IRR |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | GPT-5 | 79 | 73–86 | +3 | 0,4 | 1,3 | 1,7 | 100 % | 0,60 | 75 % |
| 2 | Claude Opus 4.7 | 71 | 63–80 | -3 | 0,6 | 2,6 | 3,2 | 100 % | 1,71 | 63 % |
| 3 | Mistral Large 2 | 52 | 39–65 | 0 | 1,3 | 2,1 | 3,3 | 100 % | 3,95 | 57 % |
| 4 | Gemini 2.5 Pro | 38 | 27–53 | +18 | 1,4 | 0,4 | 1,7 | 100 % | 4,42 | 46 % |
Raw data of this run
raw.jsonl and progress.jsonl are the public streams with response texts stripped (contamination protection, methodology §7).