TimeLitmus - Finance Natural Prediction: leaderboard
TimeLitmus (Gong et al., 2026) tests event-conditioned time-series prediction: given a textual event and the serialized pre-event series, the model predicts the expected response. In the Finance domain, built on FNSPID market news and prices, it picks one of four classes (strong or weak, upward or downward) on 380 natural records; accuracy in percent, higher is better.
Metric: Natural accuracy (%; 380 Finance records of FNSPID market events with four expected-response classes (strong or weak, up or down); ten LLMs with deterministic decoding where available, invalid or unparsable outputs scored incorrect). Source: arxiv.org. Saturation forecast: Around 2031. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5 | 51.3 | #256 |
| 2 | DeepSeek V4 Flash | 48.7 | #254 |
| 3 | MiniMax-M3 | 47.1 | #265 |
| 4 | GPT-5.4 | 44.7 | #121 |
| 5 | DeepSeek R1 0528 | 44.5 | #425 |
| 6 | Gemini 3.5 Flash | 42.1 | #98 |
| 7 | Qwen Plus | 41.1 | #788 |
| 8 | Qwen 3.5 4B | 40.8 | #1180 |
| 9 | Qwen 3.5 9B | 40.3 | #803 |
| 10 | Claude Sonnet 4.6 | 37.1 | #158 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=timelitmus-finance-natural-prediction · How It Works · Data refreshed daily, snapshot 2026-11-12.