DeepSeek V5 (Leaked Staging) vs MiniMax M3.1-Flash-Preview
A direct, empirical evaluation of DeepSeek V5 (Leaked Staging) (DeepSeek AI) against MiniMax M3.1-Flash-Preview (MiniMax).
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.
MiniMax M3.1-Flash-Preview
Developed by MiniMax · Proprietary Sparse Coding MoE parameters
Low-latency interactive IDE code completions at 165 tokens/sec, SWE-bench Verified bug fixes (73.8%), terminal automation, and budget-optimized pull request reviews.
| Metric | DeepSeek V5 (Leaked Staging) | MiniMax M3.1-Flash-Preview |
|---|---|---|
| SWE-bench Verified | 81.4% | 73.8% |
| MMLU-Pro | 87.5% | 81.2% |
| MATH-500 | 96.8% | 92.4% |
| Context Window | 1,000,000 tokens | 1,000,000 tokens |
| Input Pricing (per 1M) | $0.12 | $0.10 |
| Output Pricing (per 1M) | $0.48 | $0.40 |