DeepSeek V3.2
DeepSeek · Released Dec 2025
DeepSeek V3.2 is a large language model that prioritizes computational efficiency while maintaining reasoning and tool-use capabilities, with a 128K token context window.
- Strengths
- Uses sparse attention mechanisms to reduce computational cost compared to standard dense transformers while preserving performance on reasoning and multi-step tasks.
- Best for
- Applications requiring extended context handling and agentic tool-use where inference efficiency matters, such as document processing and autonomous workflows.
- Limitations
- As a sparse attention model, it may have reduced effectiveness on tasks requiring dense global attention patterns, and performance characteristics differ from standard dense models.
Input / 1M
$0.269
Output / 1M
$0.4
Cached input / 1M
$0.1345
Context window
163K
Price history
Input (solid)Output (dashed)
Price change
- 30d
- in increased 17.6% out increased 16.6%
- 90d
- in increased 17.6% out increased 16.6%
- 1y
- in increased 17.6% out increased 16.6%
- Since launch
- in increased 17.6% out increased 16.6%
Snapshots
| Effective | Input | Output | Cached in | Note | Source |
|---|---|---|---|---|---|
| 15 Jul 2026 | $0.269 | $0.4 | $0.1345 | Imported from OpenRouter | openrouter.ai |
| 10 Jul 2026 | $0.2145 | $0.3218 | $0.0214 | Imported from OpenRouter | openrouter.ai |
| 27 Jun 2026 | $0.2288 | $0.3432 | $0.0229 | Imported from OpenRouter | openrouter.ai |
| 11 Jun 2026 | $0.2288 | $0.3432 | — | Imported from OpenRouter | openrouter.ai |