DeepSeek V5 (Leaked Staging) vs Visual Jev (Qwen3-VL 8B Backbone)
A direct, empirical evaluation of DeepSeek V5 (Leaked Staging) (DeepSeek AI) against Visual Jev (Qwen3-VL 8B Backbone) (Independent AI Systems Lab (Guanxu Yu & Yuhang Yao)).
DeepSeek V5 (Leaked Staging)
Developed by DeepSeek AI · 1.8 Trillion (Sparse MoE) parameters
Autonomous coding agents scoring 81.4% on SWE-bench Verified, zero-telemetry self-hosting, sub-cent 1M-token processing, and formal mathematical proofs.
Visual Jev (Qwen3-VL 8B Backbone)
Developed by Independent AI Systems Lab (Guanxu Yu & Yuhang Yao) · 8.2 Billion Vision-Language Parameters parameters
Sub-20ms multi-query image verification, automated rejection sampling in generative diffusion pipelines (FLUX / SD3), robotics perception routing, and high-frequency visual software inspection.
| Metric | DeepSeek V5 (Leaked Staging) | Visual Jev (Qwen3-VL 8B Backbone) |
|---|---|---|
| SWE-bench Verified | 81.4% | 62.1% |
| MMLU-Pro | 87.5% | 86.4% |
| MATH-500 | 96.8% | 89.2% |
| Context Window | 1,000,000 tokens | 32,768 tokens (Visual prefix + batched question suffixes) |
| Input Pricing (per 1M) | $0.12 | $0.05 |
| Output Pricing (per 1M) | $0.48 | $0.10 |