Ranking run 2026-05-18
Claude Opus 4.7 leads with 79/100 across all domains. Tax law: Claude Opus 4.7 (73) · Medicine: Claude Opus 4.7 (87) · Law: Claude Opus 4.7 (83) · Business law: GPT-5 (77).
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 | 21 / 33 | 79 | 76–82 | 0 | 58 | 1,46 | — |
| 2 | GPT-5 | 0 | 31 / 53 | 75 | 72–79 | 0 | 69 | 0,62 | — |
| 3 | Gemini 2.5 Pro | 0 | 19 / 24 | 54 | 50–57 | +1 | 81 | 3,32 | — |
| 4 | Mistral Large 2 | 0 | 13 / 75 | 49 | 45–54 | -2 | 50 | 4,15 | — |
| 5 | Grok 4.3 | 0 | 11 / 16 | 46 | 42–50 | 0 | 71 | 5,08 | — |
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 0,4 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 1,3 per answer. Relative to answer length: 1,71 per 1000 output tokens, lowest GPT-5 at 0,70.
| # | 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 | 73 | 68–78 | -3 | 0,4 | 1,8 | 2,2 | 100 % | 1,71 | 78 % |
| 2 | GPT-5 | 60 | 51–69 | +1 | 1,0 | 0,8 | 1,8 | 100 % | 0,70 | 66 % |
| 3 | Gemini 2.5 Pro | 40 | 32–50 | +6 | 1,1 | 0,4 | 1,3 | 87 % | 5,91 | 59 % |
| 4 | Mistral Large 2 | 24 | 15–31 | -5 | 2,1 | 1,5 | 3,6 | 100 % | 7,54 | 53 % |
| 5 | Grok 4.3 | 22 | 14–31 | -4 | 1,6 | 0,9 | 2,5 | 87 % | 10,43 | 59 % |
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 0,4 times per answer and adds avg 1,0 claims the judge panel flags as false or unsupported; 90 % 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: 1,21 per 1000 output tokens, lowest GPT-5 at 0,90.
| # | 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 | 87 | 81–92 | +3 | 0,4 | 1,0 | 1,3 | 90 % | 1,21 | 83 % |
| 2 | GPT-5 | 82 | 73–88 | +1 | 0,4 | 0,8 | 1,2 | 95 % | 0,90 | 80 % |
| 3 | Mistral Large 2 | 64 | 56–71 | 0 | 0,4 | 0,9 | 1,2 | 90 % | 2,33 | 67 % |
| 4 | Gemini 2.5 Pro | 56 | 48–64 | 0 | 0,2 | 0,3 | 0,4 | 45 % | 2,01 | 76 % |
| 5 | Grok 4.3 | 51 | 43–58 | -1 | 0,4 | 0,4 | 0,7 | 75 % | 3,12 | 73 % |
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; 86 % of answers carry at least one flag. The lowest value is Gemini 2.5 Pro at avg 0,3 per answer. Relative to answer length: 1,13 per 1000 output tokens, lowest GPT-5 at 0,46.
| # | 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 | 79–87 | +3 | 0,2 | 0,8 | 1,0 | 86 % | 1,13 | 83 % |
| 2 | GPT-5 | 76 | 72–81 | 0 | 0,1 | 0,6 | 0,8 | 86 % | 0,46 | 72 % |
| 3 | Gemini 2.5 Pro | 63 | 57–67 | +1 | 0,2 | 0,2 | 0,3 | 43 % | 1,64 | 72 % |
| 4 | Grok 4.3 | 54 | 49–58 | +5 | 0,4 | 0,3 | 0,6 | 63 % | 4,26 | 68 % |
| 5 | Mistral Large 2 | 50 | 43–57 | +3 | 0,9 | 0,6 | 1,4 | 77 % | 4,62 | 68 % |
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 0,5 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 1,2 per answer. 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 | 77 | 70–84 | -1 | 0,5 | 0,9 | 1,4 | 100 % | 0,60 | 74 % |
| 2 | Claude Opus 4.7 | 73 | 67–79 | -2 | 0,4 | 2,0 | 2,4 | 100 % | 1,71 | 75 % |
| 3 | Mistral Large 2 | 52 | 45–59 | -5 | 0,8 | 1,5 | 2,3 | 93 % | 3,62 | 60 % |
| 4 | Gemini 2.5 Pro | 48 | 42–55 | -4 | 0,7 | 0,5 | 1,2 | 77 % | 4,30 | 59 % |
| 5 | Grok 4.3 | 45 | 38–51 | -2 | 0,6 | 0,7 | 1,3 | 90 % | 4,35 | 53 % |
Raw data of this run
raw.jsonl and progress.jsonl are the public streams with response texts stripped (contamination protection, methodology §7).