Ranking run 2026-05-10
Claude Opus 4.7 leads with 77/100 across all domains. Tax law: Claude Opus 4.7 (68) · Medicine: GPT-5 (83) · Law: Claude Opus 4.7 (82) · Business law: GPT-5 (76).
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 | 20 / 32 | 77 | 73–80 | -1 | 52 | 1,50 | — |
| 2 | GPT-5 | 0 | 34 / 46 | 67 | 59–75 | -9 | 60 | 0,74 | — |
| 3 | Mistral Large 2 | 0 | 16 / 368 | 46 | 37–55 | +1 | 38 | 4,44 | — |
| 4 | Gemini 2.5 Pro n=11/33 | 22 | 18 / 34 | 21 | 13–30 | +6 | 97 | 1,17 | — |
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
In Tax law, Claude Opus 4.7 (top score) contradicts a stored key fact on average 0,6 times per answer and adds avg 1,8 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 0,0 per answer. Relative to answer length: 1,82 per 1000 output tokens, lowest Gemini 2.5 Pro at 0,00.
| # | 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 | 68 | 63–73 | -3 | 0,6 | 1,8 | 2,5 | 100 % | 1,82 | 81 % |
| 2 | GPT-5 | 42 | 26–57 | -48 | 2,0 | 1,0 | 3,0 | 100 % | 1,11 | 70 % |
| 3 | Mistral Large 2 | 7 | 2–13 | -5 | 2,8 | 1,7 | 4,5 | 100 % | 9,96 | 56 % |
| 4 | Gemini 2.5 Pro | 0 | 0–0 | -12 | 0,0 | 0,0 | 0,0 | 0 % | 0,00 | 100 % |
Medicine · 9 Items
Ranks 1–3 statistically tied (95 % CI overlaps rank 1)
In Medicine, GPT-5 (top score) contradicts a stored key fact on average 0,4 times per answer and adds avg 0,5 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 0,2 per answer. Relative to answer length: 0,62 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 | 83 | 73–92 | +1 | 0,4 | 0,5 | 0,9 | 100 % | 0,62 | 72 % |
| 2 | Claude Opus 4.7 | 82 | 77–87 | -4 | 0,2 | 1,0 | 1,2 | 100 % | 1,19 | 83 % |
| 3 | Mistral Large 2 | 68 | 61–75 | +2 | 0,3 | 1,0 | 1,3 | 100 % | 2,20 | 75 % |
| 4 | Gemini 2.5 Pro | 29 | 19–40 | +10 | 0,0 | 0,2 | 0,2 | 40 % | 1,73 | 88 % |
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 1,1 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 0,0 per answer. Relative to answer length: 1,47 per 1000 output tokens, lowest Gemini 2.5 Pro at 0,00.
| # | 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 | 82 | 76–87 | +2 | 0,3 | 1,1 | 1,4 | 100 % | 1,47 | 76 % |
| 2 | GPT-5 | 64 | 49–78 | -1 | 0,5 | 0,5 | 1,0 | 75 % | 0,52 | 70 % |
| 3 | Mistral Large 2 | 53 | 49–56 | +4 | 0,5 | 0,9 | 1,4 | 100 % | 3,97 | 63 % |
| 4 | Gemini 2.5 Pro | 17 | 7–34 | +10 | 0,0 | 0,0 | 0,0 | 0 % | 0,00 | 78 % |
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 0,6 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 0,0 per answer. Relative to answer length: 0,59 per 1000 output tokens, lowest Gemini 2.5 Pro at 0,00.
| # | Model | top100 | 95 % CI | Δ prev. | contradicts key fact | extra false/unsupp. | avg total | answers affected | per 1000 tok. | IRR |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | GPT-5 | 76 | 69–83 | — | 0,6 | 0,9 | 1,5 | 100 % | 0,59 | 74 % |
| 2 | Claude Opus 4.7 | 74 | 65–82 | 0 | 0,6 | 2,1 | 2,7 | 100 % | 1,48 | 78 % |
| 3 | Mistral Large 2 | 52 | 42–64 | -1 | 1,1 | 1,6 | 2,7 | 100 % | 3,42 | 55 % |
| 4 | Gemini 2.5 Pro | 20 | 20–20 | — | 0,0 | 0,0 | 0,0 | 0 % | 0,00 | 80 % |
Raw data of this run
raw.jsonl and progress.jsonl are the public streams with response texts stripped (contamination protection, methodology §7).