DeepSeek-V3.2-Exp
by DeepSeek
· DeepSeek V3 family · released Sep 29, 2025
Open weights
Active
Experimental update to DeepSeek-V3.1 introducing DeepSeek Sparse Attention (DSA) to cut long-context attention cost dramatically, with API prices reduced by ~50%+. MIT-licensed weights on Hugging Face.
Pricing
Launch price (from Sep 29, 2025)
$0.280 in / $0.420 out per 1M tokens
Current price (since Sep 29, 2025)
$0.280 in / $0.420 out per 1M tokens
| Tier | Effective | In $/M | Out $/M | Cache $/M | Per image | Per second | Source |
|---|---|---|---|---|---|---|---|
| Standard | Sep 29, 2025 – now | $0.280 | $0.420 | - | - | - | - |
Specifications
| Model type | Text LLM |
|---|---|
| Architecture | MoE |
| Parameters | 685 (37 active) |
| Context window | 163.8K tokens |
| Max output | 65.5K tokens |
| Knowledge cutoff | Not disclosed |
| License | MIT License |
| Modalities | Text input Text output |
| Tags | Coding Long context Reasoning Tool use |
| End of life | Not disclosed |
| Weights | Download |
| Last verified | Not disclosed |