ParlaCAP Test Datasets
Models read parliamentary speeches and choose a policy topic or Other. The four-language test compares predictions with human labels on 3,443 speeches.
The score file covers 18 language model entries plus five classifiers or baselines. The June 2025 evaluation covered English and Croatian.
- Made by
- ParlaCAP project
- Released
- June 2025
- Kind
- Reading in many languages
- Level
- School
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- Public, plus a private test set
- Measures
- Capability
- Subject
- Political science
- Interaction
- Single turn
- Items from
- Another benchmark, Real work and use
- Refresh
- One fixed set
- Run by
- A host
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- September 2025
- Scores refreshed
- November 20th, 2026
GPT-5 leads with 0.69 among models released by September 2025 GPT-5 leads with 0.69 among models released by September 2025
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. The authors report no overall score, so this is the plain mean of macro F1 across four languages. 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 September 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
- 0.69 GPT-5
- 0.69 ParlaCAP-classifier
- 0.68 GPT-4o
- 0.68 GPT-5 mini
- 0.65
Gemini 2.5 Flash
- 0.61 GPT-5 nano
- 0.61
Gemini 2.5 Flash-Lite
- 0.60
Mistral Medium 3.1
- 0.59
Mistral Small 3.2
- 0.59 GPT-4o mini
- 0.58
Llama 3.3 70B Instruct
- 0.57
Qwen3-32B
- 0.56
Gemma 3 27B
- 0.52 GPT-3.5 Turbo
- 0.46
Llama 4 Scout
- 0.32
DeepSeek-R1-Distill-Qwen-14B
- 0.26 GaMS-27B-quantized
- 0.26 GaMS-27B
- 0.05 SVC
- 0.02 COMPLEMENTNB
- 1 GPT-5 0.69
- 2 ParlaCAP-classifier 0.69
- 3 GPT-4o Aug 2024 0.68
- 4 GPT-5 mini 0.68
- 5
Gemini 2.5 Flash Sep 2025 0.65
- 6 GPT-5 nano 0.61
- 7
Gemini 2.5 Flash-Lite Sep 2025 0.61
- 8
Mistral Medium 3.1 0.60
- 9
Mistral Small 3.2 0.59
- 10 GPT-4o mini 0.59
- 11
Llama 3.3 70B Instruct 0.58
- 12
Qwen3-32B 0.57
- 13
Gemma 3 27B 0.56
- 14 GPT-3.5 Turbo Jan 2024 0.52
- 15
Llama 4 Scout 0.46
- 16
DeepSeek-R1-Distill-Qwen-14B 0.32
- 17 GaMS-27B-quantized 0.26
- 18 GaMS-27B 0.26
- 19 SVC 0.05
- 20 COMPLEMENTNB 0.02
Source: Official ParlaCAP Test Datasets leaderboard
- 1 GPT-5 0.69
- 2 GPT-4o Aug 2024 0.68
- 3 GPT-5 mini 0.68
- 4
Gemini 2.5 Flash Sep 2025 0.65
- 5 GPT-5 nano 0.61
- 6
Gemini 2.5 Flash-Lite Sep 2025 0.61
- 7
Mistral Medium 3.1 0.60
- 8
Mistral Small 3.2 0.59
- 9 GPT-4o mini 0.59
- 10
Llama 3.3 70B Instruct 0.58
- 11
Qwen3-32B 0.57
- 12
Gemma 3 27B 0.56
- 13 GPT-3.5 Turbo Jan 2024 0.52
- 14
Llama 4 Scout 0.46
- 15
DeepSeek-R1-Distill-Qwen-14B 0.32
Source: Official ParlaCAP Test Datasets leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 0.69 | Aug 2025 | Official leaderboard | |
| 2 | ParlaCAP-classifier | 0.69 | Official leaderboard | ||
| 3 | GPT-4o Aug 2024 | 0.68 | Aug 2024 | Official leaderboard | |
| 4 | GPT-5 mini | 0.68 | Aug 2025 | Official leaderboard | |
| 5 | | 0.65 | Sep 2025 | Official leaderboard | |
| 6 | GPT-5 nano | 0.61 | Aug 2025 | Official leaderboard | |
| 7 | | 0.61 | Sep 2025 | Official leaderboard | |
| 8 | | 0.60 | Aug 2025 | Official leaderboard | |
| 9 | | 0.59 | Jun 2025 | Official leaderboard | |
| 10 | GPT-4o mini | 0.59 | Jul 2024 | Official leaderboard | |
| 11 | | 0.58 | Dec 2024 | Official leaderboard | |
| 12 | | 0.57 | Apr 2025 | Official leaderboard | |
| 13 | | 0.56 | Mar 2025 | Official leaderboard | |
| 14 | GPT-3.5 Turbo Jan 2024 | 0.52 | Jan 2024 | Official leaderboard | |
| 15 | | 0.46 | Apr 2025 | Official leaderboard | |
| 16 | | 0.32 | Jan 2025 | Official leaderboard | |
| 17 | GaMS-27B-quantized | 0.26 | Official leaderboard | ||
| 18 | GaMS-27B | 0.26 | Official leaderboard | ||
| 19 | SVC | 0.05 | Official leaderboard | ||
| 20 | COMPLEMENTNB | 0.02 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 0.69 | Aug 2025 | Official leaderboard | |
| 2 | GPT-4o Aug 2024 | 0.68 | Aug 2024 | Official leaderboard | |
| 3 | GPT-5 mini | 0.68 | Aug 2025 | Official leaderboard | |
| 4 | | 0.65 | Sep 2025 | Official leaderboard | |
| 5 | GPT-5 nano | 0.61 | Aug 2025 | Official leaderboard | |
| 6 | | 0.61 | Sep 2025 | Official leaderboard | |
| 7 | | 0.60 | Aug 2025 | Official leaderboard | |
| 8 | | 0.59 | Jun 2025 | Official leaderboard | |
| 9 | GPT-4o mini | 0.59 | Jul 2024 | Official leaderboard | |
| 10 | | 0.58 | Dec 2024 | Official leaderboard | |
| 11 | | 0.57 | Apr 2025 | Official leaderboard | |
| 12 | | 0.56 | Mar 2025 | Official leaderboard | |
| 13 | GPT-3.5 Turbo Jan 2024 | 0.52 | Jan 2024 | Official leaderboard | |
| 14 | | 0.46 | Apr 2025 | Official leaderboard | |
| 15 | | 0.32 | Jan 2025 | Official leaderboard |
Rankings by part
We read 4 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Croatian, English and Serbian. GPT-5 leads 3 of them, and GPT-4o leads the other one. Arrows show how far a model moved from its overall rank among the models we track.
Croatian
15 models we track have a score here.
- 1 GPT-5 0.69
- 2 GPT-5 mini ↑1 0.68
- 3 GPT-4o Aug 2024 ↓1 0.67
- 4
Gemini 2.5 Flash Sep 2025 0.66
- 5
Mistral Medium 3.1 ↑2 0.63
- 6 GPT-5 nano ↓1 0.61
- 7
Gemini 2.5 Flash-Lite Sep 2025 ↓1 0.60
- 8
Qwen3-32B ↑3 0.60
- 9 GPT-4o mini 0.59
- 10
Mistral Small 3.2 ↓2 0.58
English
15 models we track have a score here.
- 1 GPT-4o Aug 2024 ↑1 0.74
- 2 GPT-5 ↓1 0.73
- 3 GPT-5 mini 0.73
- 4
Gemini 2.5 Flash Sep 2025 0.71
- 5
Gemini 2.5 Flash-Lite Sep 2025 ↑1 0.68
- 6
Mistral Medium 3.1 ↑1 0.67
- 7 GPT-5 nano ↓2 0.66
- 8
Qwen3-32B ↑3 0.65
- 9 GPT-4o mini 0.64
- 10
Llama 3.3 70B Instruct 0.63
Serbian
15 models we track have a score here.
- 1 GPT-5 0.71
- 2 GPT-4o Aug 2024 0.69
- 3 GPT-5 mini 0.68
- 4
Gemini 2.5 Flash Sep 2025 0.63
- 5 GPT-5 nano 0.63
- 6
Gemini 2.5 Flash-Lite Sep 2025 0.62
- 7
Llama 3.3 70B Instruct ↑3 0.61
- 8
Mistral Small 3.2 0.60
- 9
Mistral Medium 3.1 ↓2 0.60
- 10 GPT-4o mini ↓1 0.59
Bosnian
15 models we track have a score here.
- 1 GPT-5 0.64
- 2 GPT-4o Aug 2024 0.63
- 3 GPT-5 mini 0.62
- 4
Gemini 2.5 Flash Sep 2025 0.59
- 5 GPT-5 nano 0.56
- 6
Mistral Small 3.2 ↑2 0.54
- 7
Llama 3.3 70B Instruct ↑3 0.53
- 8 GPT-4o mini ↑1 0.52
- 9
Gemini 2.5 Flash-Lite Sep 2025 ↓3 0.52
- 10
Mistral Medium 3.1 ↓3 0.51
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 ParlaCAP Test Datasets 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 ParlaCAP Test Datasets 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 Multilingual
- MMMLU In use MMLU professionally translated into 14 languages. Multiple-choice accuracy. Claude Mythos Preview 92.7%
- WMT24++ In use English-to-55-language translation of WMT24 test data. Scored with automatic metrics such as MetricX or xCOMET.
K-EXAONE 90.5%
- MMLU-ProX In use MMLU-Pro translated into up to 29 languages with expert review. Multiple-choice accuracy per language.
Qwen3.7-Max 87%
- CLIcK In use Models answer 1,995 multiple-choice questions on Korean language and culture. Questions come from official Korean exams and textbooks in eleven categories.
Solar Open 2 90.7%
- Global MMLU Lite 3.0 In use Models answer 400 human-translated MMLU questions per language in 23 languages. Version 3.0 adds Czech, Hungarian, Oriya, Slovak and Tajik and updates Italian. Accuracy. Claude Opus 5 95.7%
- MMMB In use Models answer multiple-choice questions about images in six languages across 15 categories. The set has about 12,000 questions. LFM2.5-VL-3B 83%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite. 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 2 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (20 entries).
- Pass Models scored 15 models we track, plus 5 other entries on its leaderboard.
- Warning Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 15 rows for models we track and 5 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.