TFRBench - NYC Taxi - Logic-to-Number Consistency: leaderboard

Forecasting reasoning on the hourly NYC taxi demand series of TFRBench (2026): the model reasons step by step about trends and external events before forecasting, and Gemini 3 Pro compares that reasoning with verified reference reasoning. Mean judge score from 1 to 5 for logic-to-number consistency, the agreement between the narrative and the numerical forecast; higher is better.

Metric: Mean LLM-judge score (1-5) for logic-to-number consistency of the model's chain-of-thought forecasting reasoning on the TFRBench hourly NYC taxi demand (transportation) windows, judged by Gemini 3 Pro against the benchmark's verified reference reasoning (w/ Reasoning setting: the model reasons step by step from the history window, then forecasts; no tools; mean of three runs); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro4.87#145
2Gemini 2.5 Pro4.36#276
3Gemini 2.5 Flash4.33#463
4Claude Sonnet 42.87#410
5Claude Sonnet 4.52.77#261

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

Interactive version: theaggregate.ai/benchmark?slug=tfrbench-nyc-taxi-logic-to-number-consistency · How It Works · Data refreshed daily, snapshot 2026-11-21.