AI-Roundtable Leaderboard
All runs

One official ranking run per month, permanently addressable.

One entry per run with date, panel, item count and question-bank version. Each detail page freezes the leaderboard for exactly that run — score, confidence interval, hallucination rate per domain and the diff to the previous run — and links the raw data. This is the URL to cite.

Machine-readable: /runs/latest.json · /runs/index.json · /feed.xml (Atom) · /rss.xml · per domain Steuerrecht, Medizin, Jura, Wirtschaftsrecht

Official runs (6)

DateRun-IDModelsItems Recommendation · scoreDomain winnersQuestion bank
2026-09-10run-20260910T064154-bada0b14126Claude Opus 5 · 88Tax law Claude Opus 5 Medicine Claude Opus 5 Law Claude Fable 5.1 Business law Claude Opus 52026-09-09-v5
2026-09-07run-20260907T181839-5742df14100Claude Opus 5 · 86Tax law Claude Opus 5 Medicine Claude Fable 5.1 Law Claude Opus 5 Business law Claude Opus 52026-05-17-v1
2026-08-01run-20260801T055440-4527919100Claude Opus 4.7 · 81Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-5.6 Sol2026-05-17-v1
2026-07-19run-20260719T120108-c408399100Claude Opus 4.7 · 80Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law Claude Opus 4.72026-05-17-v1
2026-06-21run-20260621T143914-12112d9100GPT-5.5 Pro · 80Tax law Claude Opus 4.7 Medicine GPT-5.5 Pro Law Claude Opus 4.7 Business law GPT-5.5 Pro2026-05-17-v1
2026-05-29run-20260529T043500-b4ffe06100Claude Opus 4.7 · 79Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-17-v1

Development runs (25)

Roster tests, interim runs after model changes and the May 2026 pilot runs. Fully archived and individually addressable, but not in the feed and not on the homepage — one official run per calendar month, so subscribers never see three conflicting rankings in one week.

DateRun-IDModelsItems Best model · scoreDomain winnersQuestion bank
2026-09-04run-20260904T151155-c623e014100Claude Opus 5 · 86Tax law Claude Opus 5 Medicine Claude Opus 5 Law Claude Opus 5 Business law Claude Opus 52026-05-17-v1
2026-09-01run-20260901T060205-7d51d09100Claude Opus 4.7 · 78Tax law Kimi K3 Medicine GPT-5.6 Sol Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-07-15run-20260715T040307-80bf679100Claude Opus 4.7 · 79Tax law Claude Opus 4.7 Medicine GPT-5.6 Sol Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-07-11run-20260711T103548-436ac19100Claude Opus 4.7 · 79Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-07-01run-20260701T043526-2a195e9100Claude Opus 4.7 · 80Tax law Claude Opus 4.7 Medicine GPT-5.5 Pro Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-06-20run-20260620T073912-d2125c8100Claude Opus 4.7 · 80Tax law Claude Opus 4.7 Medicine GPT-5.5 Pro Law Claude Opus 4.7 Business law GPT-5.5 Pro2026-05-17-v1
2026-06-19run-20260619T195640-4ce8c88100Claude Opus 4.7 · 78Tax law Claude Opus 4.7 Medicine GPT-5.5 Pro Law Claude Opus 4.7 Business law GPT-5.5 Pro2026-05-17-v1
2026-06-18run-20260618T165158-d2543a7100Claude Opus 4.7 · 80Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-06-15run-20260615T043639-30c7616100Claude Opus 4.7 · 79Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-06-07run-20260607T181746-23cbb66100Claude Opus 4.7 · 81Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-05-28run-20260528T202802-6ce0965100Claude Opus 4.7 · 79Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-05-20run-20260520T074328-4a549c5100Claude Opus 4.7 · 76Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-05-18run-20260518T043700-b4a3805100Claude Opus 4.7 · 79Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-05-17run-20260517T131406-f961285100Claude Opus 4.7 · 79Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-17-v1
2026-05-16run-20260516T200347-bc6991533Claude Opus 4.7 · 76Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-08-pilot
2026-05-16run-20260516T101429-0435c4533Claude Opus 4.7 · 78Tax law Claude Opus 4.7 Medicine GPT-5 Law Claude Opus 4.7 Business law GPT-52026-05-08-pilot
2026-05-14run-20260514T075711-8acb3a499Claude Opus 4.7 · 75Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-08-pilot
2026-05-12run-20260512T135155-40f852433Claude Opus 4.7 · 77Tax law Claude Opus 4.7 Medicine GPT-5 Law Claude Opus 4.7 Business law Claude Opus 4.72026-05-08-pilot
2026-05-12run-20260512T105829-f0ac95433Claude Opus 4.7 · 78Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-08-pilot
2026-05-12run-20260512T081552-d5fb45433Claude Opus 4.7 · 77Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-08-pilot
2026-05-11run-20260511T051320-09184c499Claude Opus 4.7 · 75Tax law Claude Opus 4.7 Medicine GPT-5 Law Claude Opus 4.7 Business law GPT-52026-05-08-pilot
2026-05-10run-20260510T190628-59fc01433Claude Opus 4.7 · 76Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law GPT-52026-05-08-pilot
2026-05-10run-20260510T175034-5e294d433Claude Opus 4.7 · 77Tax law Claude Opus 4.7 Medicine GPT-5 Law Claude Opus 4.7 Business law GPT-52026-05-08-pilot
2026-05-10run-20260510T164305-2c7b16433Claude Opus 4.7 · 78Tax law GPT-5 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law Claude Opus 4.72026-05-08-pilot
2026-05-10run-20260510T121001-e3a863433Claude Opus 4.7 · 77Tax law Claude Opus 4.7 Medicine Claude Opus 4.7 Law Claude Opus 4.7 Business law Claude Opus 4.72026-05-08-pilot

Rule for “official”: since code_version 0.2.10 the cron run sets a flag in the MANIFEST; older runs count as official when they are the last full run (≥ 50 items, non-pilot bank) of their calendar month. Runs are never altered after publication (append-only) — the one exception is a documented single-model merge, shown as “merged” on the detail page.