Ranking run 2026-05-16
Claude Opus 4.7 leads with 76/100 across all domains. Tax law: Claude Opus 4.7 (64) · Medicine: Claude Opus 4.7 (87) · Law: Claude Opus 4.7 (79) · 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 | 22 / 36 | 76 | 71–82 | -2 | 55 | 1,40 | — |
| 2 | GPT-5 | 0 | 49 / 66 | 71 | 62–78 | -2 | 61 | 0,71 | — |
| 3 | Mistral Large 2 | 0 | 23 / 366 | 45 | 37–53 | +1 | 40 | 4,33 | — |
| 4 | Gemini 2.5 Pro | 0 | 19 / 22 | 44 | 38–51 | -2 | 73 | 4,23 | — |
| 5 | Grok 4.3 | 0 | 7 / 10 | 42 | 34–51 | +1 | 60 | 5,81 | — |
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 0,8 times per answer and adds avg 1,5 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 1,8 per answer. Relative to answer length: 1,73 per 1000 output tokens, lowest GPT-5 at 0,87.
| # | 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 | 64 | 58–72 | -5 | 0,8 | 1,5 | 2,3 | 100 % | 1,73 | 77 % |
| 2 | GPT-5 | 46 | 26–65 | -15 | 1,6 | 0,8 | 2,4 | 100 % | 0,87 | 65 % |
| 3 | Gemini 2.5 Pro | 29 | 16–43 | +3 | 1,4 | 0,5 | 1,8 | 88 % | 8,36 | 64 % |
| 4 | Grok 4.3 | 18 | 4–35 | 0 | 2,4 | 1,0 | 3,3 | 100 % | 11,68 | 56 % |
| 5 | Mistral Large 2 | 13 | 8–19 | 0 | 2,0 | 1,7 | 3,7 | 100 % | 7,77 | 59 % |
Medicine · 9 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 0,2 times per answer and adds avg 0,8 claims the judge panel flags as false or unsupported; 89 % 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: 0,93 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 | 87 | 79–94 | +6 | 0,2 | 0,8 | 1,0 | 89 % | 0,93 | 88 % |
| 2 | GPT-5 | 83 | 77–89 | -2 | 0,2 | 0,8 | 1,0 | 100 % | 0,71 | 85 % |
| 3 | Mistral Large 2 | 65 | 58–71 | -3 | 0,3 | 1,0 | 1,3 | 100 % | 2,23 | 70 % |
| 4 | Grok 4.3 | 59 | 49–69 | +7 | 0,3 | 0,4 | 0,7 | 44 % | 2,94 | 73 % |
| 5 | Gemini 2.5 Pro | 52 | 41–62 | 0 | 0,3 | 0,4 | 0,6 | 67 % | 2,72 | 68 % |
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 0,0 times per answer and adds avg 1,0 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 0,2 per answer. Relative to answer length: 1,14 per 1000 output tokens, lowest GPT-5 at 0,47.
| # | 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 | 79 | 68–88 | -6 | 0,0 | 1,0 | 1,0 | 100 % | 1,14 | 76 % |
| 2 | GPT-5 | 74 | 63–84 | +6 | 0,1 | 0,7 | 0,8 | 63 % | 0,47 | 77 % |
| 3 | Gemini 2.5 Pro | 54 | 44–63 | +1 | 0,1 | 0,1 | 0,2 | 25 % | 0,99 | 75 % |
| 4 | Grok 4.3 | 50 | 42–57 | +2 | 0,3 | 0,3 | 0,4 | 63 % | 2,80 | 65 % |
| 5 | Mistral Large 2 | 47 | 38–58 | -2 | 1,0 | 0,6 | 1,5 | 88 % | 4,94 | 63 % |
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 0,5 times per answer and adds avg 1,5 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 1,7 per answer. Relative to answer length: 0,70 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 | 72–85 | -1 | 0,5 | 1,5 | 2,0 | 100 % | 0,70 | 71 % |
| 2 | Claude Opus 4.7 | 74 | 63–84 | -3 | 0,5 | 2,5 | 3,0 | 100 % | 1,59 | 64 % |
| 3 | Mistral Large 2 | 51 | 40–63 | +8 | 1,6 | 1,7 | 3,2 | 100 % | 3,82 | 54 % |
| 4 | Grok 4.3 | 41 | 24–58 | -3 | 1,3 | 0,8 | 2,0 | 100 % | 4,71 | 55 % |
| 5 | Gemini 2.5 Pro | 40 | 29–51 | -10 | 1,3 | 0,5 | 1,7 | 100 % | 4,70 | 51 % |
Raw data of this run
raw.jsonl and progress.jsonl are the public streams with response texts stripped (contamination protection, methodology §7).