TimeLitmus - Finance Event-Side Counterfactual Pairs: leaderboard
A controlled diagnostic of TimeLitmus (Gong et al., 2026): each of 80 Finance pairs keeps the price series and changes a mechanism-relevant attribute of the market event so that the expected response changes. Both endpoints are evaluated independently and a pair counts only when both predictions are correct; pair correctness in percent, higher is better.
Metric: Counterfactual pair correctness (%; both independently evaluated endpoints correct on 80 Finance pairs whose event attribute is changed while the series is kept; ten LLMs with deterministic decoding where available, invalid or unparsable outputs scored incorrect). Source: arxiv.org. Saturation forecast: Not forecast. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen Plus | 61.3 | #788 |
| 2 | Claude Sonnet 4.6 | 50 | #158 |
| 3 | Qwen 3.5 9B | 38.8 | #803 |
| 4 | Gemini 3.5 Flash | 37.5 | #98 |
| 5 | MiniMax-M3 | 25 | #265 |
| 6 | DeepSeek R1 0528 | 18.8 | #425 |
| 7 | DeepSeek V4 Flash | 17.5 | #254 |
| 8 | GLM-5 | 16.2 | #256 |
| 9 | Qwen 3.5 4B | 13.8 | #1180 |
| 10 | GPT-5.4 | 5 | #121 |
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-event-side-counterfactual-pairs · How It Works · Data refreshed daily, snapshot 2026-11-12.