TimeLitmus - Finance Hard Paired Contrasts: leaderboard
Hard paired contrasts in TimeLitmus (Gong et al., 2026): 360 naturally occurring pairs of similar Finance event and price inputs that require different responses, each endpoint evaluated independently. A pair counts only when both predictions are correct; pair correctness in percent, higher is better.
Metric: Hard paired contrast pair correctness (%; both endpoints correct on 360 naturally occurring similar Finance pairs with different gold responses; 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 | GLM-5 | 19.2 | #256 |
| 2 | DeepSeek V4 Flash | 16.7 | #254 |
| 3 | DeepSeek R1 0528 | 16.1 | #425 |
| 4 | Claude Sonnet 4.6 | 15 | #158 |
| 5 | MiniMax-M3 | 13.9 | #265 |
| 6 | Qwen 3.5 4B | 12.2 | #1180 |
| 7 | Gemini 3.5 Flash | 9.4 | #98 |
| 8 | Qwen Plus | 8.9 | #788 |
| 9 | GPT-5.4 | 8.6 | #121 |
| 10 | Qwen 3.5 9B | 3.3 | #803 |
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-hard-paired-contrasts · How It Works · Data refreshed daily, snapshot 2026-11-12.