DeepSeek-V4-Pro vs Grok 4
Direct spec comparison of DeepSeek-V4-Pro (from DeepSeek) and Grok 4 (from xAI). Want a 3- or 4-way comparison? Open the multi-model tool →
 DeepSeek-V4-ProDeepSeek |  Grok 4xAI | |
|---|---|---|
| Vendor | DeepSeek | xAI |
| Family | DeepSeek | Grok |
| Release date | 2026-04-22 | 2025-07-09 |
| Context window | 1,000,000 tokens | 256,000 tokens |
| Parameters | 1.6T (49B active) | — |
| Modality | text | text |
| License | MIT | proprietary |
| Source | open weights | proprietary |
| Description | DeepSeek's flagship open-weight MoE. 1.6T parameters with 49B activated, 1M-token context, and a hybrid attention scheme (CSA + HCA) that delivers long-context inference at ~27% of V3.2's FLOPs. | — |
| Links |
More like this
Looking for a different head-to-head? Build your own comparison on the multi-model tool.
Or see all DeepSeek-V4-Pro details / Grok 4 details.