DeepSeek V5 (Leaked Staging) vs Gemini 4 Carbon
A direct, empirical evaluation of DeepSeek V5 (Leaked Staging) (DeepSeek AI) against Gemini 4 Carbon (Google DeepMind).
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.
Gemini 4 Carbon
Developed by Google DeepMind · Carbon-Aware Frontier Transformer on TPU v6e parameters
Sub-watt green computing, 62% energy reduction per million tokens, carbon-aware datagrid routing, and eco-tier batch execution discounts.
| Metric | DeepSeek V5 (Leaked Staging) | Gemini 4 Carbon |
|---|---|---|
| SWE-bench Verified | 81.4% | 69.2% |
| MMLU-Pro | 87.5% | 84.6% |
| MATH-500 | 96.8% | 91.4% |
| Context Window | 1,000,000 tokens | 1,000,000 tokens |
| Input Pricing (per 1M) | $0.12 | $0.80 |
| Output Pricing (per 1M) | $0.48 | $2.40 |