DeepSeek-V4-Pro-0813 vs Gemini 3 Pro
DeepSeek-V4-Pro-0813 (DeepSeek) and Gemini 3 Pro (Google DeepMind) compared on benchmarks, pricing, context window, and use-case rankings.
- DeepSeek-V4-Pro-0813 leads on overall intelligence (ECI 155 vs 153).
- DeepSeek-V4-Pro-0813 ranks higher for coding (78th vs 43th percentile).
| Attribute | DeepSeek-V4-Pro-0813 | Gemini 3 Pro |
|---|---|---|
| Scores | ||
| Intelligence (ECI) | 155 | 153 |
| Coding | 78 | 43 |
| Math | 88 | 66 |
| Reasoning & Knowledge | 69 | 68 |
| Agentic & Tools | — | 49 |
| Specifications | ||
| Developer | DeepSeek | Google DeepMind |
| Family | DeepSeek | Gemini |
| Released | Aug 13, 2026 | Nov 18, 2025 |
| Parameters | 1.6T | — |
| Availability | Open weights (unrestricted) | API access |
| Context window | 1M | — |
| Price — $/M input | $0.58 | — |
| Price — $/M output | $1.74 | — |
| Inputs | text | — |
| Outputs | text | — |
| Benchmarks | ||
| AIME 2024/2025 | 99% | 91% |
| ARC-AGI | 91% | 75% |
| ARC-AGI-2 | 61% | 31% |
| GPQA Diamond | 92% | 93% |
| Humanity's Last Exam | 41% | 38% |
| SciCode | 51% | — |
| SimpleQA Verified | 53% | 73% |
| WebDev Arena | 1582 | 1438 |
| WeirdML | 66% | 70% |
| APEX | — | 32% |
| FrontierMath | — | 38% |
| FrontierMath Tier 4 | — | 19% |
| GDPval (win/tie rate) | — | 54% |
| GSO (code optimization) | — | 19% |
| METR task horizon | — | 3.7 h |
| SimpleBench | — | 76% |
| SWE-bench Verified | — | 73% |
| Terminal-Bench | — | 69% |
Use-case scores are 0–100 percentile composites across each area’s benchmarks, ranked against every model from the past year. Highlighted cells lead each row. Open a model for the full picture.
Frequently asked questions
Is DeepSeek-V4-Pro-0813 better than Gemini 3 Pro?
On Epoch AI's Capabilities Index, DeepSeek-V4-Pro-0813 scores higher (155) than Gemini 3 Pro (153). The right pick depends on your task — compare their coding, math, and reasoning scores in the table above.
Which is better for coding, DeepSeek-V4-Pro-0813 or Gemini 3 Pro?
Across coding benchmarks like SWE-bench Verified and Terminal-Bench, DeepSeek-V4-Pro-0813 ranks higher — 78th vs 43th percentile among the models tracked on Model Beat.
Want a different match-up? Open the compare tool to add or swap models.
More comparisons
Benchmarks & model data from Epoch AI (CC BY); pricing & specs from OpenRouter. ECI = Epoch Capabilities Index.