LiveNewsBench (Nov-Dec 2025): leaderboard
Metric: Accuracy (%) on LiveNewsBench's 200 human-verified questions about news events of November and December 2025, local ReAct agent with at most 5 search queries and 5 page visits per question, answers graded by GPT-4.1 with the SimpleQA grading prompt; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3.2 (Thinking) | 84.5 | #198 (DeepSeek V3.2) |
| 2 | DeepSeek V3.2 (Non-reasoning) | 83 | #198 (DeepSeek V3.2) |
| 3 | Claude Sonnet 4.5 | 82 | #138 |
| 4 | Grok 4 | 82 | #169 |
| 5 | GPT-5.2 | 74 | #105 |
| 6 | GPT-4.1 | 72.5 | #240 |
| 7 | Qwen 3 235B A22B 2507 Instruct | 61.5 | #291 |
| 8 | Gemini 3 Pro | 60.5 | #77 |
| 9 | Qwen 3 235B A22B 2507 (Thinking) | 60 | #253 (Qwen 3 235B A22B 2507) |
| 10 | Qwen 3 8B (Non-reasoning) | 49.5 | #667 (Qwen 3 8B) |
| 11 | Kimi K2 (Thinking) | 48 | #236 (Kimi K2) |
| 12 | GPT-OSS-120B | 27 | #330 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=livenewsbench-nov-dec-2025 · How It Works · Data refreshed daily, snapshot 2026-10-11.