Fox1.3 Benchmark Results
Performance Metrics
Speed Benchmark
| Test | Tokens | Time | Speed |
|---|---|---|---|
| Capital of France | 105 | 1.97s | 53.3 tok/s |
| Photosynthesis | 107 | 1.70s | 62.8 tok/s |
| Math problem (3x+15=45) | 114 | 1.68s | 67.7 tok/s |
| Python code (prime check) | 109 | 1.70s | 64.3 tok/s |
| AI vs Machine Learning | 110 | 1.76s | 62.4 tok/s |
Average Speed: 61.8 tokens/sec
Comparison with Claude Opus 4.5
| Metric | Fox1.3 | Claude Opus 4.5 |
|---|---|---|
| Parameters | ~850M | ~175B |
| Speed | 61.8 tok/s | ~20 tok/s |
| Size | 1.7 GB | ~350 GB |
| Time to first token | ~50ms | ~1840ms |
| Context window | 32K | 200K |
| Running cost | Free (local) | $5-$25/1M tokens |
Note: Fox1.3 is ~3x faster than Opus 4.5, but Opus is more intelligent due to 200x more parameters.
Hardware
- GPU: NVIDIA RTX 3060
- Memory: 12GB VRAM
Model Details
- Base: Qwen2.5-1B-Instruct
- Training: LoRA fine-tuning on 150+ examples
- Trained: 2026-04-01
- Dataset includes: general knowledge, science, math, reasoning, coding
Links
- Model: https://huggingface.co/teolm30/fox1.3
- Training data: Available in repository