HeaRTS - Trajectory Analysis: leaderboard
Metric: Mean task score (x100) over HeaRTS's trajectory analysis tasks (Deduction category: analyzing health trajectories across visits or sessions), each scored 0 to 1 by accuracy; the model reasons over the signal files by writing and running Python code in a CodeAct agent loop with a minimal package set; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 Mini | 55 | #346 |
| 2 | MiniMax-M2 | 55 | #307 |
| 3 | Nemotron Nano 12B V2 | 54 | #656 |
| 4 | DeepSeek V3.1 | 52 | #260 |
| 5 | GLM-5 (Thinking) | 52 | #137 (GLM-5) |
| 6 | Claude Haiku 4.5 | 50 | #271 |
| 7 | Llama 4 Maverick | 49 | #451 |
| 8 | GPT-5 Mini | 48 | #176 |
| 9 | Qwen 3 Coder 480B A35B Instruct | 47 | #302 |
| 10 | GLM-4.7 (Thinking) | 47 | #185 (GLM-4.7) |
| 11 | Gemini 2.5 Flash | 46 | #237 |
| 12 | Gemini 2.5 Pro | 45 | #145 |
| 13 | Grok 4.1 Fast (Reasoning) | 45 | #208 (Grok 4.1 Fast) |
| 14 | Kimi K2 (Thinking) | 43 | #236 (Kimi K2) |
| 15 | Gemini 3.1 Pro (Preview) | 39 | #54 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=hearts-trajectory-analysis · How It Works · Data refreshed daily, snapshot 2026-10-11.