Ranking run 2026-06-07
Claude Opus 4.7 leads with 81/100 across all domains. Tax law: Claude Opus 4.7 (77) · Medicine: Claude Opus 4.7 (86) · Law: Claude Opus 4.7 (84) · Business law: GPT-5 (83).
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 / 30 | 81 | 78–83 | +2 | 59 | 1,48 | — |
| 2 | GPT-5 | 0 | 53 / 146 | 77 | 74–81 | +2 | 72 | 0,57 | — |
| 3 | Claude Opus 4.8 | 0 | 18 / 26 | 74 | 71–77 | — | 72 | 0,93 | — |
| 4 | Gemini 2.5 Pro | 0 | 18 / 24 | 54 | 50–57 | +2 | 79 | 3,46 | — |
| 5 | Mistral Large 2 | 0 | 10 / 68 | 50 | 46–55 | 0 | 49 | 4,32 | — |
| 6 | Grok 4.3 | 0 | 4 / 6 | 47 | 43–50 | 0 | 73 | 5,50 | — |
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 1–3 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 2,3 claims the judge panel flags as false or unsupported; 100 % of answers carry at least one flag. The lowest value is Claude Opus 4.8 at avg 1,4 per answer. Relative to answer length: 2,10 per 1000 output tokens, lowest GPT-5 at 0,71.
| # | 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 | 71–82 | +1 | 0,4 | 2,3 | 2,7 | 100 % | 2,10 | 78 % |
| 2 | Claude Opus 4.8 | 69 | 63–76 | — | 0,6 | 0,9 | 1,4 | 93 % | 1,04 | 73 % |
| 3 | GPT-5 | 67 | 58–75 | +6 | 0,9 | 1,0 | 1,9 | 100 % | 0,71 | 70 % |
| 4 | Gemini 2.5 Pro | 35 | 26–44 | +2 | 1,3 | 0,5 | 1,7 | 80 % | 7,60 | 62 % |
| 5 | Grok 4.3 | 26 | 19–33 | -7 | 1,5 | 0,7 | 2,1 | 100 % | 9,44 | 54 % |
| 6 | Mistral Large 2 | 25 | 17–32 | -1 | 2,1 | 1,0 | 3,0 | 100 % | 6,34 | 49 % |
Medicine · 20 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 0,4 times per answer and adds avg 0,8 claims the judge panel flags as false or unsupported; 95 % of answers carry at least one flag. The lowest value is Gemini 2.5 Pro at avg 0,6 per answer. Relative to answer length: 1,17 per 1000 output tokens, lowest Claude Opus 4.8 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 | 86 | 80–91 | -1 | 0,4 | 0,8 | 1,1 | 95 % | 1,17 | 80 % |
| 2 | GPT-5 | 82 | 75–88 | +3 | 0,4 | 0,5 | 0,9 | 100 % | 0,66 | 75 % |
| 3 | Claude Opus 4.8 | 80 | 74–87 | — | 0,3 | 0,4 | 0,6 | 85 % | 0,57 | 75 % |
| 4 | Mistral Large 2 | 66 | 59–72 | 0 | 0,5 | 0,7 | 1,1 | 90 % | 2,21 | 72 % |
| 5 | Gemini 2.5 Pro | 57 | 49–64 | +1 | 0,3 | 0,4 | 0,6 | 65 % | 2,72 | 71 % |
| 6 | Grok 4.3 | 53 | 46–61 | -2 | 0,5 | 0,2 | 0,7 | 75 % | 3,49 | 70 % |
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 0,3 times per answer and adds avg 0,7 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: 1,10 per 1000 output tokens, lowest GPT-5 at 0,53.
| # | 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 | 80–87 | +2 | 0,3 | 0,7 | 0,9 | 86 % | 1,10 | 80 % |
| 2 | Claude Opus 4.8 | 79 | 72–84 | — | 0,3 | 0,5 | 0,8 | 63 % | 0,85 | 76 % |
| 3 | GPT-5 | 74 | 68–80 | 0 | 0,3 | 0,6 | 0,9 | 80 % | 0,53 | 75 % |
| 4 | Gemini 2.5 Pro | 62 | 58–67 | +1 | 0,2 | 0,2 | 0,3 | 40 % | 1,36 | 69 % |
| 5 | Grok 4.3 | 48 | 42–54 | 0 | 0,6 | 0,2 | 0,8 | 63 % | 6,09 | 66 % |
| 6 | Mistral Large 2 | 47 | 39–54 | -1 | 1,2 | 0,8 | 2,0 | 83 % | 6,71 | 64 % |
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 0,2 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. Relative to answer length: 0,49 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 | 77–88 | +5 | 0,2 | 0,9 | 1,1 | 100 % | 0,49 | 78 % |
| 2 | Claude Opus 4.7 | 76 | 70–81 | +4 | 0,3 | 2,0 | 2,3 | 97 % | 1,60 | 70 % |
| 3 | Claude Opus 4.8 | 67 | 61–72 | — | 0,6 | 1,1 | 1,7 | 93 % | 1,11 | 65 % |
| 4 | Mistral Large 2 | 57 | 50–64 | +3 | 0,7 | 1,5 | 2,1 | 97 % | 3,37 | 54 % |
| 5 | Gemini 2.5 Pro | 51 | 44–57 | +3 | 0,7 | 0,5 | 1,1 | 73 % | 3,97 | 59 % |
| 6 | Grok 4.3 | 51 | 45–57 | +4 | 0,4 | 0,8 | 1,2 | 83 % | 4,50 | 60 % |
Raw data of this run
raw.jsonl and progress.jsonl are the public streams with response texts stripped (contamination protection, methodology §7).