Ranking run 2026-06-15
Claude Opus 4.7 leads with 79/100 across all domains. Tax law: Claude Opus 4.7 (75) · Medicine: Claude Opus 4.7 (85) · Law: Claude Opus 4.7 (80) · Business law: GPT-5 (80).
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 | 76–82 | -2 | 56 | 1,59 | — |
| 2 | Claude Opus 4.8 | 0 | 17 / 29 | 74 | 71–77 | 0 | 73 | 0,89 | — |
| 2 | GPT-5 | 0 | 23 / 42 | 74 | 70–78 | -3 | 69 | 0,65 | — |
| 3 | Gemini 2.5 Pro | 0 | 19 / 23 | 54 | 50–58 | 0 | 82 | 3,17 | — |
| 4 | Mistral Large 2 | 0 | 10 / 66 | 49 | 44–54 | -1 | 47 | 4,46 | — |
| 5 | Grok 4.3 | 0 | 8 / 10 | 48 | 44–52 | +1 | 76 | 4,58 | — |
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 0,3 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 1,5 per answer. Relative to answer length: 1,76 per 1000 output tokens, lowest GPT-5 at 0,75.
| # | 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 | 75 | 68–81 | -2 | 0,3 | 2,0 | 2,3 | 100 % | 1,76 | 68 % |
| 2 | Claude Opus 4.8 | 66 | 60–73 | -3 | 0,5 | 1,1 | 1,6 | 100 % | 1,20 | 73 % |
| 3 | GPT-5 | 59 | 46–70 | -8 | 1,1 | 0,9 | 2,0 | 100 % | 0,75 | 55 % |
| 4 | Gemini 2.5 Pro | 38 | 27–48 | +3 | 1,1 | 0,6 | 1,5 | 93 % | 6,88 | 53 % |
| 5 | Mistral Large 2 | 30 | 20–41 | +5 | 1,5 | 2,0 | 3,5 | 100 % | 7,29 | 57 % |
| 6 | Grok 4.3 | 26 | 17–36 | 0 | 1,7 | 1,0 | 2,7 | 100 % | 11,37 | 58 % |
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 0,5 times per answer and adds avg 0,8 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 0,2 per answer. Relative to answer length: 1,21 per 1000 output tokens, lowest Claude Opus 4.8 at 0,65.
| # | 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 | 85 | 80–90 | -1 | 0,5 | 0,8 | 1,2 | 100 % | 1,21 | 75 % |
| 2 | GPT-5 | 80 | 73–86 | -2 | 0,5 | 0,4 | 0,9 | 80 % | 0,66 | 77 % |
| 3 | Claude Opus 4.8 | 79 | 71–87 | -1 | 0,3 | 0,4 | 0,7 | 75 % | 0,65 | 68 % |
| 4 | Mistral Large 2 | 69 | 63–75 | +3 | 0,4 | 0,7 | 1,1 | 95 % | 2,18 | 74 % |
| 5 | Gemini 2.5 Pro | 61 | 54–68 | +4 | 0,1 | 0,1 | 0,2 | 40 % | 0,98 | 70 % |
| 6 | Grok 4.3 | 52 | 47–58 | -1 | 0,3 | 0,2 | 0,4 | 55 % | 2,05 | 75 % |
Law · 35 Items
Ranks 1–3 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; 74 % of answers carry at least one flag. The lowest value is Grok 4.3 at avg 0,4 per answer. Relative to answer length: 1,26 per 1000 output tokens, lowest Claude Opus 4.8 at 0,62.
| # | 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 | 75–85 | -4 | 0,3 | 0,7 | 1,0 | 74 % | 1,26 | 81 % |
| 2 | Claude Opus 4.8 | 80 | 75–84 | +1 | 0,2 | 0,4 | 0,6 | 66 % | 0,62 | 77 % |
| 3 | GPT-5 | 73 | 65–79 | -1 | 0,5 | 0,6 | 1,1 | 80 % | 0,69 | 74 % |
| 4 | Gemini 2.5 Pro | 59 | 54–64 | -3 | 0,2 | 0,2 | 0,4 | 40 % | 1,93 | 71 % |
| 5 | Grok 4.3 | 51 | 46–55 | +3 | 0,3 | 0,2 | 0,4 | 40 % | 2,83 | 64 % |
| 6 | Mistral Large 2 | 42 | 34–49 | -5 | 1,3 | 0,7 | 1,9 | 80 % | 6,32 | 66 % |
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 0,4 times per answer and adds avg 0,9 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 1,0 per answer. Relative to answer length: 0,54 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 | 80 | 75–85 | -3 | 0,4 | 0,9 | 1,2 | 90 % | 0,54 | 68 % |
| 2 | Claude Opus 4.7 | 77 | 72–82 | +1 | 0,7 | 2,0 | 2,7 | 100 % | 1,92 | 72 % |
| 3 | Claude Opus 4.8 | 68 | 62–74 | +1 | 0,5 | 1,1 | 1,6 | 93 % | 1,06 | 59 % |
| 4 | Mistral Large 2 | 54 | 48–60 | -3 | 0,8 | 1,5 | 2,3 | 97 % | 3,62 | 52 % |
| 5 | Grok 4.3 | 53 | 48–58 | +2 | 0,5 | 0,6 | 1,1 | 90 % | 3,81 | 61 % |
| 6 | Gemini 2.5 Pro | 52 | 47–58 | +1 | 0,6 | 0,5 | 1,0 | 80 % | 3,75 | 61 % |
Raw data of this run
raw.jsonl and progress.jsonl are the public streams with response texts stripped (contamination protection, methodology §7).