RT-QA - Direct (L1): leaderboard

RT-QA (2026) asks real-time questions in Chinese whose answers come from executing a verified scraping workflow at the moment of inference; models run as search agents with web search and page reading on six days in January 2026, judged by GPT-5.2. This board covers direct questions that need no entity resolution. Accuracy in percent; higher is better.

Metric: Accuracy (%) on direct (0-hop) questions, mean of the six days; Chinese real-time questions whose answers come from executing a verified scraping workflow at the moment of inference, asked daily from January 10 to 15, 2026; a DeepResearch-style agent with Serper search and Jina page reading, the current timestamp in the prompt, temperature 0.6; a GPT-5.2 judge checks semantic equivalence; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GLM-4.746.9#380
2Seed 1.844.3#252
3RT-QA Claude-4.5 (checkpoint unspecified)44.3
4GPT-5.243.2#185
5RT-QA Kimi-K2 (checkpoint unspecified)43
6Gemini 3 Pro41#145
7DeepSeek V3.234.2#419
8RT-QA Qwen3-235B (checkpoint unspecified)33

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=rt-qa-direct-l1 · How It Works · Data refreshed daily, snapshot 2026-11-21.