Açaí vs Qwen
Qwen (通义千问) is Alibaba's flagship multimodal model, and it's a genuine competitor to GPT-4 class systems, especially for Mandarin and regional cuisine recognition. That still does not make it a nutrition tracker.
What Qwen does well vs what a tracker requires
| Capability | Açaí | Qwen |
|---|---|---|
| Recognizes Chinese regional dishes | Yes, trained on global cuisine | Excellent, a clear Qwen strength |
| Recognizes Western / global cuisine | Yes | Very good |
| Returns structured 245-nutrient JSON | Native schema | Requires prompt engineering |
| Estimates portion sizes | Benchmarked at 93 kcal MAE | Guesses verbally |
| Persists your meal log | Unlimited history | Chat history only, no semantic recall |
| Real-time BMR and deficit | Yes, live | No body model |
| Projects end-of-day landing | Yes | Only on request |
| Weekly deficiency heatmap | Yes | No, stateless |
| Apple Health integration | Yes | No |
| Google Fit integration | Yes | No |
| Available in Chinese | UI in English, food recognition global | Native Chinese, strong |
| Works inside a phone app | iOS + Android app | Browser + API mostly |
Strong model, still not a product
Qwen is arguably the best open-weights class model for Chinese-language food recognition. If your diet includes a lot of regional cuisine, asking Qwen about a photo will often return more useful detail than asking GPT or Gemini the same question. That's a real strength.
What Qwen can't do, by design, is the rest of what a nutrition tracker does: persistent per-user storage, a structured 245-field schema, a live BMR and deficit model, health integration, and a UI designed for daily capture. Those are product decisions, not model decisions.
Açaí handles the food recognition, including common Chinese dishes, and wraps it in the product surface Qwen can't. The combination a power user would ideally want: use Açaí for the daily practice, use Qwen for a specific regional food question when Açaí's confidence is low.
Turn AI capability into a daily result.
Açaí gives Qwen-class intelligence a persistent memory and a real health model.