CzechTopic - Text-Level F1: leaderboard
Metric: Text-level F1 (%) for deciding whether the topic is present in the text, on CzechTopic's 1,820 human-annotated (text, topic) pairs (525 historical Czech texts, 363 topics defined by a name and description), zero-shot or two-shot prompting with the authors' span tagging or matching prompts, macro-averaged over topics against the human annotators' spans; higher is better. Source: arxiv.org. Saturation forecast: Around November 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 80.6 | #105 |
| 2 | Llama 3.3 70B Instruct | 75.2 | #520 |
| 3 | GPT-OSS-20B | 74.7 | #499 |
| 4 | Gemma 3 27B | 70 | #596 |
| 5 | GPT-5 Mini (2025-08-07) | 67.8 | #165 |
| 6 | Gemini 3 Pro (Preview) | 62.4 | #64 |
| 7 | Gemma 3 4B | 59.9 | #1084 |
| 8 | Llama 3.2 3B Instruct | 59.7 | #1321 |
| 9 | GPT-5 Nano | 18.4 | #415 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=czechtopic-text-level-f1 · How It Works · Data refreshed daily, snapshot 2026-10-11.