ThermEval - Thermal Reasoning (Two People): leaderboard
Metric: Accuracy (times 100, 0-100) on Task 5, whether the chest, forehead or nose of the left or the right person is hotter, from a ThermEval-D image with a colorbar (155 questions); zero-shot with fixed one-word or one-number prompts, a single forward pass; Gemini 2.5 models parse free-form answers into the label or number only when the output deviates from the format; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 74 | #77 |
| 2 | Llama 3.2 11B Instruct | 61 | #1112 |
| 3 | Gemini 2.5 Pro | 57 | #145 |
| 4 | Gemini 3 Flash | 55 | #93 |
| 5 | InternVL3-14B | 53 | #494 |
| 6 | Gemini 2.5 Flash | 52 | #237 |
| 7 | InternVL3-8B | 48 | #606 |
| 8 | InternVL3-38B | 48 | #395 |
| 9 | Qwen 2.5 VL 7B Instruct | 44 | #643 |
| 10 | MiniCPM-V-2.6 | 41 | #825 |
| 11 | Qwen 2 VL 7B Instruct | 41 | #816 |
| 12 | Qwen 2.5 VL 32B Instruct | 37 | #443 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=thermeval-thermal-reasoning-two-people · How It Works · Data refreshed daily, snapshot 2026-10-11.