TopBench (Agentic) - Single-Point Prediction: leaderboard
TopBench (2026) tests implicit predictive reasoning in table question answering: the answer is not in the table and must be predicted from historical rows. Prediction accuracy from 0 to 1 on single-point questions, solved by an agent that runs Python in a sandbox; higher is better.
Metric: Prediction accuracy (0 to 1) on TopBench's 274 single-point questions, which ask for an unobserved outcome of a profile given a historical table, solved by the authors' ReAct agent that writes and runs Python in a Docker sandbox over the table: a regression answer earns 0.6 x (1 - range-normalized absolute error) plus 0.4 x the overlap of its predicted interval with the reference interval, penalized when the interval is too wide, and a classification answer earns credit for the right class; values and decisions extracted from the free-text answer by a DeepSeek-V3.2 judge and checked against the answer text; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3 Flash | 0.66 | #175 | |
| 2 | Claude Sonnet 4.5 | 0.64 | #261 | |
| 3 | DeepSeek V3.2 | Thinking | 0.61 | #419 |
| 4 | GPT-5.2 | 0.6 | #185 | |
| 5 | Qwen 3 235B A22B 2507 (Thinking) | 0.57 | #466 | |
| 6 | Qwen 3 235B A22B 2507 Instruct | 0.43 | #616 | |
| 7 | TableLLM-13B | 0.12 | ||
| 8 | TableLLM-8B | 0.1 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=topbench-agentic-single-point-prediction · How It Works · Data refreshed daily, snapshot 2026-11-21.