GPBam
Models write legal opinions on German public law cases and retrieve the statutes they need. Three judges grade essays against reference solutions.
- Made by
- Abdullah Al Zubaer and colleagues
- Released
- November 2026
- Kind
- Law
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- Public, plus a private test set
- Measures
- Capability
- Subject
- Law
- Language
- German
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- A host
- In views
- Law and finance
- Strongest model tested
- Claude Opus 5 #11 on the capability index
- Last model added
- August 2026
- Scores refreshed
- November 20th, 2026
Claude Opus 5 leads with 76.4 among models released by August 2026 Claude Opus 5 leads with 76.4 among models released by August 2026
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in August 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 15 more
- 76.4 Claude Opus 5
- 67.1
Gemini 3.7 Flash
- 47.9
DeepSeek-V4-Flash
- 41.5
Qwen3.5 397B-A17B
- 37.8
Mistral Large 3
- 37.4
Mistral Medium 3.5
- 36.8
DeepSeek-V3.2
- 36.3 RedHatAI/gemma-4-31B-it-FP8-block
- 35.2
Gemini 2.5 Flash-Lite
- 34.8 GPT-5 mini
- 34.8 Claude Haiku 4.5
- 33.2
DeepSeek-R1
- 33.1
Qwen 3.6 35B-A3B
- 33.0
Qwen3.5-122B-A10B
- 32.6
Qwen3-Next-80B-A3B-Instruct
- 32.4 gpt-oss-120b
- 31.1
Qwen3-235B-A22B-Thinking
- 26.4
DeepSeek-V3
- 26.1 soofi-s-isar-preview
- 24.4
Llama 4 Maverick
- 23.3 GaleneAI/Magistral-Small-2509-FP8-Dynamic
- 22.9 RedHatAI/Mistral-Small-3.2-24B-Instruct-2506-FP8
- 21.5 GPT-5 nano
- 19.5 GPT-4o mini
- 17.8 windprak/open_steuerllm
- 16.8
Llama 3.3 70B Instruct
- 13.2
Qwen3-32B
- 9.26 utter-project/EuroLLM-22B-Instruct-2512
- 8.77
Ministral 3 14B
- 7.53 ibm-granite/granite-4.1-3b
- 5.43
Phi-4 Mini
- 4.81 meta-llama/llama-3.1-8b-instruct
- 1 Claude Opus 5 No retrieval 76.4
- 2
Gemini 3.7 Flash No retrieval 67.1
- 3
DeepSeek-V4-Flash cite+LS+RRF k=50 47.9
- 4
Qwen3.5 397B-A17B No retrieval 41.5
- 5
Mistral Large 3 No retrieval 37.8
- 6
Mistral Medium 3.5 No retrieval 37.4
- 7
DeepSeek-V3.2 No retrieval 36.8
- 8 RedHatAI/gemma-4-31B-it-FP8-block cite+RRF 36.3
- 9
Gemini 2.5 Flash-Lite Sep 2025, No retrieval 35.2
- 10 GPT-5 mini No retrieval 34.8
- 11 Claude Haiku 4.5 No retrieval 34.8
- 12
DeepSeek-R1 May 2025, No retrieval 33.2
- 13
Qwen 3.6 35B-A3B cite+RRF k=50 33.1
- 14
Qwen3.5-122B-A10B No retrieval 33.0
- 15
Qwen3-Next-80B-A3B-Instruct cite-only 32.6
- 16 gpt-oss-120b cite-only 32.4
- 17
Qwen3-235B-A22B-Thinking Jul 2025, No retrieval 31.1
- 18
DeepSeek-V3 Mar 2025, No retrieval 26.4
- 19 soofi-s-isar-preview No retrieval 26.1
- 20
Llama 4 Maverick No retrieval 24.4
- 21 GaleneAI/Magistral-Small-2509-FP8-Dynamic No retrieval 23.3
- 22 RedHatAI/Mistral-Small-3.2-24B-Instruct-2506-FP8 LS text 22.9
- 23 GPT-5 nano No retrieval 21.5
- 24 GPT-4o mini No retrieval 19.5
- 25 windprak/open_steuerllm No retrieval 17.8
- 26
Llama 3.3 70B Instruct No retrieval 16.8
- 27
Qwen3-32B No retrieval 13.2
- 28 utter-project/EuroLLM-22B-Instruct-2512 No retrieval 9.26
- 29
Ministral 3 14B No retrieval 8.77
- 30 ibm-granite/granite-4.1-3b No retrieval 7.53
- 31
Phi-4 Mini No retrieval 5.43
- 32 meta-llama/llama-3.1-8b-instruct No retrieval 4.81
Source: The GPBam paper
- 76.4 Claude Opus 5
- 67.1
Gemini 3.7 Flash
- 47.9
DeepSeek-V4-Flash
- 41.5
Qwen3.5 397B-A17B
- 37.8
Mistral Large 3
- 37.4
Mistral Medium 3.5
- 36.8
DeepSeek-V3.2
- 35.2
Gemini 2.5 Flash-Lite
- 34.8 GPT-5 mini
- 34.8 Claude Haiku 4.5
- 33.2
DeepSeek-R1
- 33.1
Qwen 3.6 35B-A3B
- 33.0
Qwen3.5-122B-A10B
- 32.6
Qwen3-Next-80B-A3B-Instruct
- 32.4 gpt-oss-120b
- 31.1
Qwen3-235B-A22B-Thinking
- 26.4
DeepSeek-V3
- 24.4
Llama 4 Maverick
- 21.5 GPT-5 nano
- 19.5 GPT-4o mini
- 16.8
Llama 3.3 70B Instruct
- 13.2
Qwen3-32B
- 8.77
Ministral 3 14B
- 5.43
Phi-4 Mini
- 1 Claude Opus 5 No retrieval 76.4
- 2
Gemini 3.7 Flash No retrieval 67.1
- 3
DeepSeek-V4-Flash cite+LS+RRF k=50 47.9
- 4
Qwen3.5 397B-A17B No retrieval 41.5
- 5
Mistral Large 3 No retrieval 37.8
- 6
Mistral Medium 3.5 No retrieval 37.4
- 7
DeepSeek-V3.2 No retrieval 36.8
- 8
Gemini 2.5 Flash-Lite Sep 2025, No retrieval 35.2
- 9 GPT-5 mini No retrieval 34.8
- 10 Claude Haiku 4.5 No retrieval 34.8
- 11
DeepSeek-R1 May 2025, No retrieval 33.2
- 12
Qwen 3.6 35B-A3B cite+RRF k=50 33.1
- 13
Qwen3.5-122B-A10B No retrieval 33.0
- 14
Qwen3-Next-80B-A3B-Instruct cite-only 32.6
- 15 gpt-oss-120b cite-only 32.4
- 16
Qwen3-235B-A22B-Thinking Jul 2025, No retrieval 31.1
- 17
DeepSeek-V3 Mar 2025, No retrieval 26.4
- 18
Llama 4 Maverick No retrieval 24.4
- 19 GPT-5 nano No retrieval 21.5
- 20 GPT-4o mini No retrieval 19.5
- 21
Llama 3.3 70B Instruct No retrieval 16.8
- 22
Qwen3-32B No retrieval 13.2
- 23
Ministral 3 14B No retrieval 8.77
- 24
Phi-4 Mini No retrieval 5.43
Source: The GPBam paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | 76.4 | No retrieval, Claude Code | Jul 2026 | Official leaderboard |
| 2 | | 67.1 | No retrieval | Aug 2026 | Official leaderboard |
| 3 | | 47.9 | cite+LS+RRF k=50 | Apr 2026 | Official leaderboard |
| 4 | | 41.5 | No retrieval | Feb 2026 | Official leaderboard |
| 5 | | 37.8 | No retrieval | Dec 2025 | Official leaderboard |
| 6 | | 37.4 | No retrieval | Apr 2026 | Official leaderboard |
| 7 | | 36.8 | No retrieval | Dec 2025 | Official leaderboard |
| 8 | RedHatAI/gemma-4-31B-it-FP8-block | 36.3 | cite+RRF | Official leaderboard | |
| 9 | | 35.2 | No retrieval | Sep 2025 | Official leaderboard |
| 10 | GPT-5 mini | 34.8 | No retrieval | Aug 2025 | Official leaderboard |
| 11 | Claude Haiku 4.5 | 34.8 | No retrieval | Oct 2025 | Official leaderboard |
| 12 | | 33.2 | No retrieval | May 2025 | Official leaderboard |
| 13 | | 33.1 | cite+RRF k=50 | Apr 2026 | Official leaderboard |
| 14 | | 33.0 | No retrieval | Feb 2026 | Official leaderboard |
| 15 | | 32.6 | cite-only | Sep 2025 | Official leaderboard |
| 16 | gpt-oss-120b | 32.4 | cite-only | Aug 2025 | Official leaderboard |
| 17 | | 31.1 | No retrieval | Jul 2025 | Official leaderboard |
| 18 | | 26.4 | No retrieval | Mar 2025 | Official leaderboard |
| 19 | soofi-s-isar-preview | 26.1 | No retrieval | Official leaderboard | |
| 20 | | 24.4 | No retrieval | Apr 2025 | Official leaderboard |
| 21 | GaleneAI/Magistral-Small-2509-FP8-Dynamic | 23.3 | No retrieval | Official leaderboard | |
| 22 | RedHatAI/Mistral-Small-3.2-24B-Instruct-2506-FP8 | 22.9 | LS text | Official leaderboard | |
| 23 | GPT-5 nano | 21.5 | No retrieval | Aug 2025 | Official leaderboard |
| 24 | GPT-4o mini | 19.5 | No retrieval | Jul 2024 | Official leaderboard |
| 25 | windprak/open_steuerllm | 17.8 | No retrieval | Official leaderboard | |
| 26 | | 16.8 | No retrieval | Dec 2024 | Official leaderboard |
| 27 | | 13.2 | No retrieval | Apr 2025 | Official leaderboard |
| 28 | utter-project/EuroLLM-22B-Instruct-2512 | 9.26 | No retrieval | Official leaderboard | |
| 29 | | 8.77 | No retrieval | Dec 2025 | Official leaderboard |
| 30 | ibm-granite/granite-4.1-3b | 7.53 | No retrieval | Official leaderboard | |
| 31 | | 5.43 | No retrieval | Feb 2025 | Official leaderboard |
| 32 | meta-llama/llama-3.1-8b-instruct | 4.81 | No retrieval | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | 76.4 | No retrieval, Claude Code | Jul 2026 | Official leaderboard |
| 2 | | 67.1 | No retrieval | Aug 2026 | Official leaderboard |
| 3 | | 47.9 | cite+LS+RRF k=50 | Apr 2026 | Official leaderboard |
| 4 | | 41.5 | No retrieval | Feb 2026 | Official leaderboard |
| 5 | | 37.8 | No retrieval | Dec 2025 | Official leaderboard |
| 6 | | 37.4 | No retrieval | Apr 2026 | Official leaderboard |
| 7 | | 36.8 | No retrieval | Dec 2025 | Official leaderboard |
| 8 | | 35.2 | No retrieval | Sep 2025 | Official leaderboard |
| 9 | GPT-5 mini | 34.8 | No retrieval | Aug 2025 | Official leaderboard |
| 10 | Claude Haiku 4.5 | 34.8 | No retrieval | Oct 2025 | Official leaderboard |
| 11 | | 33.2 | No retrieval | May 2025 | Official leaderboard |
| 12 | | 33.1 | cite+RRF k=50 | Apr 2026 | Official leaderboard |
| 13 | | 33.0 | No retrieval | Feb 2026 | Official leaderboard |
| 14 | | 32.6 | cite-only | Sep 2025 | Official leaderboard |
| 15 | gpt-oss-120b | 32.4 | cite-only | Aug 2025 | Official leaderboard |
| 16 | | 31.1 | No retrieval | Jul 2025 | Official leaderboard |
| 17 | | 26.4 | No retrieval | Mar 2025 | Official leaderboard |
| 18 | | 24.4 | No retrieval | Apr 2025 | Official leaderboard |
| 19 | GPT-5 nano | 21.5 | No retrieval | Aug 2025 | Official leaderboard |
| 20 | GPT-4o mini | 19.5 | No retrieval | Jul 2024 | Official leaderboard |
| 21 | | 16.8 | No retrieval | Dec 2024 | Official leaderboard |
| 22 | | 13.2 | No retrieval | Apr 2025 | Official leaderboard |
| 23 | | 8.77 | No retrieval | Dec 2025 | Official leaderboard |
| 24 | | 5.43 | No retrieval | Feb 2025 | Official leaderboard |
The best score, 76.4, costs $63 for the full run
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: The GPBam paper
| Model | Effort | Harness | Score | Cost of the full run | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 5 | No retrieval | Claude Code | 76.4 | $63 | |
| No retrieval | 67.1 | $2.12 | |||
| cite+LS+RRF k=50 | 47.9 | ||||
| RRF | 45.1 | ||||
| RRF k=50 | 44.8 | ||||
| cite-only | 44.4 | ||||
| rerank | 43.8 | ||||
| No retrieval | 43.8 | ||||
| cite+RRF k=50 | 43.1 | ||||
| LS text | 43.0 | ||||
| cite+RRF | 42.8 | ||||
| No retrieval | 41.5 | $2.33 | |||
| No retrieval | 37.8 | $0.60 | |||
| No retrieval | 37.4 | $10 | |||
| No retrieval | 36.8 | $0.26 | |||
| No retrieval | 35.2 | $0.51 | |||
| GPT-5 mini | No retrieval | 34.8 | $3.32 | ||
| Claude Haiku 4.5 | No retrieval | 34.8 | $8.23 | ||
| No retrieval | 33.2 | $1.14 | |||
| cite+RRF k=50 | 33.1 | ||||
| cite+LS+RRF k=50 | 33.1 | ||||
| No retrieval | 33.0 | $1.77 | |||
| cite-only | 33.0 | ||||
| rerank | 32.7 | ||||
| cite-only | 32.6 | ||||
| RRF k=50 | 32.6 | ||||
| gpt-oss-120b | cite-only | 32.4 | |||
| gpt-oss-120b | cite+RRF | 32.4 | |||
| LS text | 32.0 | ||||
| cite+RRF | 31.6 | ||||
| RRF | 31.5 | ||||
| cite+LS+RRF k=50 | 31.2 | ||||
| No retrieval | 31.1 | $1.75 | |||
| cite+RRF | 31.1 | ||||
| LS text | 31.0 | ||||
| gpt-oss-120b | cite+LS+RRF k=50 | 30.6 | |||
| No retrieval | 30.6 | $0.53 | |||
| cite+RRF k=50 | 30.6 | ||||
| No retrieval | 30.6 | $0.75 | |||
| gpt-oss-120b | cite+RRF k=50 | 30.5 | |||
| gpt-oss-120b | LS text | 30.3 | |||
| gpt-oss-120b | No retrieval | 29.8 | $0.19 | ||
| rerank | 29.7 | ||||
| RRF k=50 | 29.5 | ||||
| gpt-oss-120b | RRF | 29.3 | |||
| gpt-oss-120b | rerank | 28.9 | |||
| RRF | 28.5 | ||||
| gpt-oss-120b | RRF k=50 | 28.5 | |||
| No retrieval | 26.4 | $0.18 | |||
| No retrieval | 24.4 | $0.16 | |||
| GPT-5 nano | No retrieval | 21.5 | $0.43 | ||
| GPT-4o mini | No retrieval | 19.5 | $0.09 | ||
| No retrieval | 16.8 | $0.09 | |||
| No retrieval | 13.2 | $0.15 | |||
| No retrieval | 8.77 | $0.12 | |||
| No retrieval | 5.43 |
The best score, 76.4, costs $63 for the full run
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: The GPBam paper
| Model | Effort | Harness | Score | Cost of the full run | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 5 | No retrieval | Claude Code | 76.4 | $63 | |
| No retrieval | 67.1 | $2.12 | |||
| cite+LS+RRF k=50 | 47.9 | ||||
| RRF | 45.1 | ||||
| RRF k=50 | 44.8 | ||||
| cite-only | 44.4 | ||||
| rerank | 43.8 | ||||
| No retrieval | 43.8 | ||||
| cite+RRF k=50 | 43.1 | ||||
| LS text | 43.0 | ||||
| cite+RRF | 42.8 | ||||
| No retrieval | 41.5 | $2.33 | |||
| No retrieval | 37.8 | $0.60 | |||
| No retrieval | 37.4 | $10 | |||
| No retrieval | 36.8 | $0.26 | |||
| No retrieval | 35.2 | $0.51 | |||
| GPT-5 mini | No retrieval | 34.8 | $3.32 | ||
| Claude Haiku 4.5 | No retrieval | 34.8 | $8.23 | ||
| No retrieval | 33.2 | $1.14 | |||
| cite+RRF k=50 | 33.1 | ||||
| cite+LS+RRF k=50 | 33.1 | ||||
| No retrieval | 33.0 | $1.77 | |||
| cite-only | 33.0 | ||||
| rerank | 32.7 | ||||
| cite-only | 32.6 | ||||
| RRF k=50 | 32.6 | ||||
| gpt-oss-120b | cite-only | 32.4 | |||
| gpt-oss-120b | cite+RRF | 32.4 | |||
| LS text | 32.0 | ||||
| cite+RRF | 31.6 | ||||
| RRF | 31.5 | ||||
| cite+LS+RRF k=50 | 31.2 | ||||
| No retrieval | 31.1 | $1.75 | |||
| cite+RRF | 31.1 | ||||
| LS text | 31.0 | ||||
| gpt-oss-120b | cite+LS+RRF k=50 | 30.6 | |||
| No retrieval | 30.6 | $0.53 | |||
| cite+RRF k=50 | 30.6 | ||||
| No retrieval | 30.6 | $0.75 | |||
| gpt-oss-120b | cite+RRF k=50 | 30.5 | |||
| gpt-oss-120b | LS text | 30.3 | |||
| gpt-oss-120b | No retrieval | 29.8 | $0.19 | ||
| rerank | 29.7 | ||||
| RRF k=50 | 29.5 | ||||
| gpt-oss-120b | RRF | 29.3 | |||
| gpt-oss-120b | rerank | 28.9 | |||
| RRF | 28.5 | ||||
| gpt-oss-120b | RRF k=50 | 28.5 | |||
| No retrieval | 26.4 | $0.18 | |||
| No retrieval | 24.4 | $0.16 | |||
| GPT-5 nano | No retrieval | 21.5 | $0.43 | ||
| GPT-4o mini | No retrieval | 19.5 | $0.09 | ||
| No retrieval | 16.8 | $0.09 | |||
| No retrieval | 13.2 | $0.15 | |||
| No retrieval | 8.77 | $0.12 | |||
| No retrieval | 5.43 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The GPBam paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official GPBam leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Law, finance and medicine
- MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy.
Qwen 3.8 Max 80.4%
- HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
- HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
- FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy.
Ling-3.1-flash 81.7%
- KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. o1-preview 92.7%
- Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%
Data audit
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access Public tasks, plus a private test set.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (32 entries).
- Pass Models scored 24 models we track, plus 8 other entries on its leaderboard.
- Pass Newest score The newest is from August 2026, for Gemini 3.7 Flash.
- Note Frontier models None of the top 10 has a score. The strongest tested is Claude Opus 5 (#11).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 56 rows for models we track and 8 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.