Compare models
Pick up to 4models. Specs render side-by-side. Share the URL — it's stateless.
 Grok 4xAI |  Qwen3 235BAlibaba |  DeepSeek-V4-ProDeepSeek | |
|---|---|---|---|
| Vendor | xAI | Alibaba | DeepSeek |
| Family | Grok | Qwen | DeepSeek |
| Release date | 2025-07-09 | 2025-04-29 | 2026-04-22 |
| Context window | 256,000 tokens | 128,000 tokens | 1,000,000 tokens |
| Parameters | — | 235B | 1.6T (49B active) |
| Modality | text | text | text |
| License | proprietary | Apache-2.0 | MIT |
| Source | proprietary | open weights | open weights |
| Description | — | Predecessor to the Qwen3.6 family. | 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 |
