Ranking run 2026-05-17
Claude Opus 4.7 leads with 79/100 across all domains. Tax law: Claude Opus 4.7 (76) · Medicine: Claude Opus 4.7 (84) · Law: Claude Opus 4.7 (80) · Business law: GPT-5 (78).
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 | 19 / 28 | 79 | 75–82 | +3 | 59 | 1,50 | — |
| 2 | GPT-5 | 0 | 28 / 49 | 75 | 71–78 | +4 | 67 | 0,67 | — |
| 3 | Gemini 2.5 Pro | 0 | 18 / 22 | 53 | 50–57 | +9 | 79 | 3,66 | — |
| 4 | Mistral Large 2 | 0 | 18 / 79 | 51 | 46–55 | +6 | 51 | 4,20 | — |
| 5 | Grok 4.3 | 0 | 9 / 15 | 46 | 42–50 | +4 | 76 | 4,48 | — |
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 · 15 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 0,3 times per answer and adds avg 2,0 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 1,7 per answer. Relative to answer length: 1,75 per 1000 output tokens, lowest GPT-5 at 0,79.
| # | 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 | 76 | 68–84 | +12 | 0,3 | 2,0 | 2,3 | 100 % | 1,75 | 72 % |
| 2 | GPT-5 | 59 | 47–71 | +13 | 0,9 | 1,2 | 2,1 | 93 % | 0,79 | 65 % |
| 3 | Gemini 2.5 Pro | 34 | 25–44 | +5 | 1,3 | 0,6 | 1,7 | 87 % | 8,27 | 62 % |
| 4 | Mistral Large 2 | 29 | 20–37 | +16 | 1,6 | 1,4 | 3,0 | 100 % | 6,28 | 53 % |
| 5 | Grok 4.3 | 26 | 17–35 | +8 | 1,3 | 0,7 | 2,0 | 80 % | 8,65 | 61 % |
Medicine · 20 Items
Ranks 1–2 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 0,9 claims the judge panel flags as false or unsupported; 100 % of answers carry at least one flag. The lowest value is Grok 4.3 at avg 0,2 per answer. Relative to answer length: 1,25 per 1000 output tokens, lowest GPT-5 at 0,94.
| # | 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 | 84 | 77–89 | -3 | 0,3 | 0,9 | 1,2 | 100 % | 1,25 | 84 % |
| 2 | GPT-5 | 81 | 73–87 | -2 | 0,6 | 0,7 | 1,2 | 95 % | 0,94 | 80 % |
| 3 | Mistral Large 2 | 64 | 58–69 | -1 | 0,3 | 0,9 | 1,2 | 95 % | 2,29 | 68 % |
| 4 | Gemini 2.5 Pro | 56 | 47–63 | +4 | 0,2 | 0,3 | 0,5 | 45 % | 2,37 | 77 % |
| 5 | Grok 4.3 | 52 | 46–59 | -7 | 0,1 | 0,2 | 0,2 | 35 % | 1,04 | 75 % |
Law · 35 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 0,2 times per answer and adds avg 0,8 claims the judge panel flags as false or unsupported; 77 % 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: 1,14 per 1000 output tokens, lowest GPT-5 at 0,57.
| # | 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 | 80 | 74–86 | +1 | 0,2 | 0,8 | 1,0 | 77 % | 1,14 | 79 % |
| 2 | GPT-5 | 76 | 69–81 | +2 | 0,2 | 0,8 | 1,0 | 80 % | 0,57 | 71 % |
| 3 | Gemini 2.5 Pro | 62 | 56–65 | +8 | 0,3 | 0,1 | 0,4 | 29 % | 1,78 | 70 % |
| 4 | Grok 4.3 | 49 | 44–54 | -1 | 0,6 | 0,2 | 0,7 | 60 % | 4,82 | 66 % |
| 5 | Mistral Large 2 | 47 | 39–54 | 0 | 1,1 | 0,7 | 1,7 | 83 % | 5,67 | 65 % |
Business law · 30 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 0,3 times per answer and adds avg 1,0 claims the judge panel flags as false or unsupported; 97 % of answers carry at least one flag. The lowest value is Gemini 2.5 Pro at avg 1,2 per answer. Relative to answer length: 0,58 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 | 78 | 72–83 | 0 | 0,3 | 1,0 | 1,4 | 97 % | 0,58 | 70 % |
| 2 | Claude Opus 4.7 | 75 | 69–80 | +1 | 0,5 | 1,9 | 2,4 | 100 % | 1,75 | 72 % |
| 3 | Mistral Large 2 | 57 | 51–64 | +6 | 0,7 | 1,5 | 2,3 | 93 % | 3,60 | 59 % |
| 4 | Gemini 2.5 Pro | 52 | 46–59 | +12 | 0,6 | 0,6 | 1,2 | 73 % | 4,11 | 63 % |
| 5 | Grok 4.3 | 47 | 40–54 | +6 | 0,6 | 0,6 | 1,2 | 70 % | 4,22 | 59 % |
Raw data of this run
raw.jsonl and progress.jsonl are the public streams with response texts stripped (contamination protection, methodology §7).