FrenchNews-7 (Unseen Outlets): leaderboard
FrenchNews-7 (Sobhy, 2026) classifies French news articles into seven editorial desks derived from publisher URL structure. This board scores zero-shot prompting on a balanced pool of 2,100 articles from four outlets never seen in training, as mean per-desk recall in percent, equal to accuracy on the balanced pool; higher is better.
Metric: Mean per-class recall (%; seven editorial desks, Politique, International, Culture & Loisirs, Sport, Sciences & Technologies, Société and Économie, on a balanced 2,100-article pool of 300 per desk from four outlets never seen in training (Libération, BFMTV, Le Nouvel Obs, Franceinfo), gold labels from publisher URL slugs; equals accuracy on the balanced pool; zero-shot prompt with one-sentence desk definitions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 70B Instruct | 77.8 | #1357 |
| 2 | Mistral Small 3.2 | 77.2 | #1591 |
| 3 | GPT-OSS-120B | 75.3 | #700 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=frenchnews-7-unseen-outlets · How It Works · Data refreshed daily, snapshot 2026-11-13.