LifeAgentBench - Context Prompting: leaderboard
LifeAgentBench (Tian et al., EMNLP 2026) asks questions over long-horizon diet, sleep, physical-activity and emotion records of 100 AI4FoodDB participants, with answers computed by executable queries and scored by exact match (numeric answers within a small tolerance). In context prompting the user's pre-filtered records are placed in the prompt (4,096-token context, 32 answer tokens) and only single-user questions are asked. Accuracy in percent; higher is better.
Metric: Answer Accuracy (%; single-user questions, pre-filtered records in the prompt, 32 new tokens). Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Pro | 58.89 | #181 |
| 2 | GPT-4o | 57.02 | #741 |
| 3 | Qwen 3.5 9B | 45.55 | #803 |
| 4 | Gemini 2.5 Flash Lite | 44.81 | #958 |
| 5 | Claude 3 Haiku | 35.3 | #2629 |
| 6 | Gemma 2 9B (IT) | 24.44 | #2275 |
| 7 | Llama 3.1 8B Instruct | 20.65 | #4277 |
| 8 | Phi-3.5-mini-instruct | 20.57 | #5757 |
| 9 | Llama 3.2 3B Instruct | 20.18 | #7185 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=lifeagentbench-context-prompting · How It Works · Data refreshed daily, snapshot 2026-11-12.