ExpertLongBench
Models write documents for 11 professional tasks. A judge compares facts in each document with a reference checklist and scores precision and recall.
The authors rank 13 models by average F1 across 11 tasks. They release seven tasks and keep four private.
- Made by
- University of Michigan and collaborators
- Released
- June 2025
- Kind
- Tasks from real jobs
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- Public, plus a private test set
- Measures
- Capability
- Language
- English
- In and out
- Text, Molecules and sequences → Text
- Interaction
- Single turn
- Items from
- Written by experts, Real work and use, Another benchmark
- Refresh
- One fixed set
- Run by
- A host
- Occupations
- Legal, Educational instruction and library, Healthcare practitioners and technical
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- August 2025
- Scores refreshed
- November 20th, 2026
Gemini 2.5 Pro leads with 33.4% among models released by August 2025 Gemini 2.5 Pro leads with 33.4% among models released by August 2025
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. 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 2025, 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 16 more
- 1
Gemini 2.5 Pro Jun 2025 33.4%
- 2 GPT-5 31%
- 3
Gemini 2.0 Flash Feb 2025 26.8%
- 4 GPT-4o Mar 2025 26.5%
- 5 GPT-4o mini 26.1%
- 6
Llama 3.3 70B Instruct 24.6%
- 7 Mistral-Large-Instruct 24%
- 8 Mistral-Nemo-Instruct 23.8%
- 9
Qwen2.5-72B 23.8%
- 10 Llama3.1-8B-Instruct 23.4%
- 11 Claude 3.7 Sonnet 23.2%
- 12
Qwen2.5-7B 20.5%
- 13 Claude 3.5 Haiku 19.3%
Source: Official ExpertLongBench leaderboard
- 1
Gemini 2.5 Pro Jun 2025 33.4%
- 2 GPT-5 31%
- 3
Gemini 2.0 Flash Feb 2025 26.8%
- 4 GPT-4o Mar 2025 26.5%
- 5 GPT-4o mini 26.1%
- 6
Llama 3.3 70B Instruct 24.6%
- 7
Qwen2.5-72B 23.8%
- 8 Claude 3.7 Sonnet 23.2%
- 9
Qwen2.5-7B 20.5%
- 10 Claude 3.5 Haiku 19.3%
Source: Official ExpertLongBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 33.4% | Jun 2025 | Official leaderboard | |
| 2 | GPT-5 | 31% | Aug 2025 | Official leaderboard | |
| 3 | | 26.8% | Feb 2025 | Official leaderboard | |
| 4 | GPT-4o Mar 2025 | 26.5% | Mar 2025 | Official leaderboard | |
| 5 | GPT-4o mini | 26.1% | Jul 2024 | Official leaderboard | |
| 6 | | 24.6% | Dec 2024 | Official leaderboard | |
| 7 | Mistral-Large-Instruct | 24% | Official leaderboard | ||
| 8 | Mistral-Nemo-Instruct | 23.8% | Official leaderboard | ||
| 9 | | 23.8% | Sep 2024 | Official leaderboard | |
| 10 | Llama3.1-8B-Instruct | 23.4% | Official leaderboard | ||
| 11 | Claude 3.7 Sonnet | 23.2% | Feb 2025 | Official leaderboard | |
| 12 | | 20.5% | Sep 2024 | Official leaderboard | |
| 13 | Claude 3.5 Haiku | 19.3% | Oct 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 33.4% | Jun 2025 | Official leaderboard | |
| 2 | GPT-5 | 31% | Aug 2025 | Official leaderboard | |
| 3 | | 26.8% | Feb 2025 | Official leaderboard | |
| 4 | GPT-4o Mar 2025 | 26.5% | Mar 2025 | Official leaderboard | |
| 5 | GPT-4o mini | 26.1% | Jul 2024 | Official leaderboard | |
| 6 | | 24.6% | Dec 2024 | Official leaderboard | |
| 7 | | 23.8% | Sep 2024 | Official leaderboard | |
| 8 | Claude 3.7 Sonnet | 23.2% | Feb 2025 | Official leaderboard | |
| 9 | | 20.5% | Sep 2024 | Official leaderboard | |
| 10 | Claude 3.5 Haiku | 19.3% | Oct 2024 | Official leaderboard |
Rankings by part
We read 11 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as T1, T2 and T3. GPT-5 leads 5 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
T1
10 models we track have a score here.
- 1 GPT-5 ↑1 27.2%
- 2
Gemini 2.5 Pro Jun 2025 ↓1 25.4%
- 3 GPT-4o mini ↑2 16.5%
- 4 GPT-4o Mar 2025 13.2%
- 5
Gemini 2.0 Flash Feb 2025 ↓2 13.1%
- 6
Qwen2.5-72B ↑1 12.3%
- 7
Llama 3.3 70B Instruct ↓1 12.1%
- 8 Claude 3.7 Sonnet 11.5%
- 9
Qwen2.5-7B 10.7%
- 10 Claude 3.5 Haiku 2.80%
T2
10 models we track have a score here.
- 1 GPT-5 ↑1 10.3%
- 2
Gemini 2.5 Pro Jun 2025 ↓1 10%
- 3
Gemini 2.0 Flash Feb 2025 7.90%
- 4 GPT-4o Mar 2025 6.20%
- 5 GPT-4o mini 5.90%
- 6
Llama 3.3 70B Instruct 4.90%
- 7
Qwen2.5-72B 4.10%
- 8
Qwen2.5-7B ↑1 4.00%
- 9 Claude 3.5 Haiku ↑1 1.10%
- 10 Claude 3.7 Sonnet ↓2 0.90%
T3
10 models we track have a score here.
- 1 Claude 3.5 Haiku ↑9 22.1%
- 2
Gemini 2.0 Flash Feb 2025 ↑1 19.5%
- 3 Claude 3.7 Sonnet ↑5 18.1%
- 4
Qwen2.5-72B ↑3 17.4%
- 5
Gemini 2.5 Pro Jun 2025 ↓4 16.9%
- 6
Llama 3.3 70B Instruct 16.8%
- 7
Qwen2.5-7B ↑2 16.5%
- 8 GPT-4o mini ↓3 15.2%
- 9 GPT-4o Mar 2025 ↓5 15.2%
- 10 GPT-5 ↓8 5.60%
T4
10 models we track have a score here.
- 1 GPT-4o Mar 2025 ↑3 56.9%
- 2
Qwen2.5-72B ↑5 51%
- 3
Gemini 2.5 Pro Jun 2025 ↓2 50.2%
- 4
Llama 3.3 70B Instruct ↑2 49.5%
- 5
Gemini 2.0 Flash Feb 2025 ↓2 49.3%
- 6 GPT-4o mini ↓1 49.2%
- 7 GPT-5 ↓5 47.2%
- 8
Qwen2.5-7B ↑1 42.5%
- 9 Claude 3.7 Sonnet ↓1 30.4%
- 10 Claude 3.5 Haiku 30.3%
T5
10 models we track have a score here.
- 1 GPT-5 ↑1 56.5%
- 2
Gemini 2.5 Pro Jun 2025 ↓1 47.9%
- 3 Claude 3.7 Sonnet ↑5 35%
- 4
Gemini 2.0 Flash Feb 2025 ↓1 30.2%
- 5 GPT-4o Mar 2025 ↓1 29.9%
- 6 GPT-4o mini ↓1 25%
- 7
Llama 3.3 70B Instruct ↓1 15.9%
- 8
Qwen2.5-7B ↑1 12.7%
- 9
Qwen2.5-72B ↓2 10.8%
- 10 Claude 3.5 Haiku 9.70%
T6
10 models we track have a score here.
- 1 GPT-5 ↑1 54.7%
- 2
Gemini 2.5 Pro Jun 2025 ↓1 44%
- 3 Claude 3.7 Sonnet ↑5 26.1%
- 4 GPT-4o mini ↑1 25.5%
- 5 GPT-4o Mar 2025 ↓1 25.3%
- 6
Qwen2.5-7B ↑3 22%
- 7
Qwen2.5-72B 21.2%
- 8
Gemini 2.0 Flash Feb 2025 ↓5 20.8%
- 9
Llama 3.3 70B Instruct ↓3 20.2%
- 10 Claude 3.5 Haiku 10.9%
T7
10 models we track have a score here.
- 1
Gemini 2.5 Pro Jun 2025 50.4%
- 2 GPT-5 49.8%
- 3 Claude 3.7 Sonnet ↑5 36.1%
- 4
Gemini 2.0 Flash Feb 2025 ↓1 35.6%
- 5
Qwen2.5-7B ↑4 34.8%
- 6 GPT-4o Mar 2025 ↓2 34.5%
- 7 GPT-4o mini ↓2 33.9%
- 8 Claude 3.5 Haiku ↑2 33.4%
- 9
Llama 3.3 70B Instruct ↓3 33.4%
- 10
Qwen2.5-72B ↓3 32.6%
T8
10 models we track have a score here.
- 1 GPT-5 ↑1 15.3%
- 2
Gemini 2.5 Pro Jun 2025 ↓1 14.9%
- 3 GPT-4o mini ↑2 10.3%
- 4 GPT-4o Mar 2025 10.1%
- 5
Qwen2.5-72B ↑2 9.30%
- 6 Claude 3.5 Haiku ↑4 9.20%
- 7 Claude 3.7 Sonnet ↑1 9.20%
- 8
Gemini 2.0 Flash Feb 2025 ↓5 9.00%
- 9
Qwen2.5-7B 8.70%
- 10
Llama 3.3 70B Instruct ↓4 8.30%
T9
10 models we track have a score here.
- 1
Gemini 2.0 Flash Feb 2025 ↑2 46%
- 2
Gemini 2.5 Pro Jun 2025 ↓1 43.2%
- 3
Qwen2.5-72B ↑4 35.7%
- 4 GPT-4o Mar 2025 35.2%
- 5
Llama 3.3 70B Instruct ↑1 32.8%
- 6 GPT-4o mini ↓1 29.5%
- 7 Claude 3.7 Sonnet ↑1 21.4%
- 8 GPT-5 ↓6 18.7%
- 9 Claude 3.5 Haiku ↑1 18.1%
- 10
Qwen2.5-7B ↓1 15.1%
T10
10 models we track have a score here.
- 1
Llama 3.3 70B Instruct ↑5 42.6%
- 2 GPT-4o mini ↑3 42.2%
- 3 Claude 3.5 Haiku ↑7 40.8%
- 4
Gemini 2.0 Flash Feb 2025 ↓1 36.7%
- 5 GPT-4o Mar 2025 ↓1 35.7%
- 6
Qwen2.5-7B ↑3 35%
- 7
Qwen2.5-72B 33.3%
- 8 Claude 3.7 Sonnet 33.1%
- 9
Gemini 2.5 Pro Jun 2025 ↓8 21.1%
- 10 GPT-5 ↓8 13.9%
T11
10 models we track have a score here.
- 1
Gemini 2.5 Pro Jun 2025 43.3%
- 2 GPT-5 41.5%
- 3 GPT-4o mini ↑2 34.2%
- 4 Claude 3.5 Haiku ↑6 34.1%
- 5
Qwen2.5-72B ↑2 33.9%
- 6
Llama 3.3 70B Instruct 33.8%
- 7 Claude 3.7 Sonnet ↑1 33.5%
- 8 GPT-4o Mar 2025 ↓4 29.2%
- 9
Gemini 2.0 Flash Feb 2025 ↓6 26.5%
- 10
Qwen2.5-7B ↓1 23.8%
2 models have held number one on the official leaderboard since its release in June 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official ExpertLongBench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official ExpertLongBench 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 Economic work
- JobBench In use 130 tasks from 35 occupations workers most want delegated. Agent works through messy reference files, graded by fact-anchored binary rubrics. GLM-5.3 61.4%
- GDPval In use Real work tasks from 44 occupations with reference files. Experts blindly compare model deliverables with expert work. Metric is win-or-tie rate.
Seed2.1 Pro 87.9%
- GDPval-AA v2.1 In use Agents with shell and web access complete 220 GDPval gold tasks from 44 occupations. Blind pairwise comparisons give Elo ratings. Claude Opus 5.5 1,846
- AA-Briefcase v1.1 In use Version 1.1 of Artificial Analysis's multi-hour office work evaluation, with Elo ratings fitted by Crowd-BT. Claude Opus 5.5 1,822
- Agents' Last Exam v1.1 In use Agents complete 122 public long-horizon professional workflow tasks across 55 sub-industries. Deterministic graders check outcomes against hidden references. GPT-6.2 Astra 47.5%
- CoWorkBench In use Agent completes long-horizon productivity tasks across computer science, finance, law and medicine. Qwen in-house scoring (method unpublished). Claude Fable 5 75.9%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 15 months ago, in August 2025, for GPT-5. Every benchmark
Catalog record 2 pass
- Pass Task access Public tasks, plus a private test set.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the official leaderboard (13 entries).
- Pass Models scored 10 models we track, plus 3 other entries on its leaderboard.
- Warning Newest score The newest came 15 months ago, in August 2025, for GPT-5.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
- Pass Headroom The best is 33.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on huggingface.co: 10 rows for models we track and 3 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.