Model specification · DeepSeek
DeepSeek V3.2
deepseek/deepseek-v3.2
DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...
SpecificationsReported values
- Context window
- 164K
- Maximum output
- 66K
- Architecture
- text->text
- Tokenizer
- DeepSeek
- Knowledge cutoff
- —
- Moderated
- No
- Input modalities
- text
- Output modalities
- text
Full pricingUSD where reported
| Price | Standard |
|---|---|
| Input | $0.269 / M |
| Output | $0.400 / M |
Benchmark indexesReported indexes
- Intelligence
- 32
- Coding
- 44.2
- Agentic
- 18.3
- Design Arena Elo
- 1167
Example workload costsStandard pricing only
| Workload | Input / output | Cache share | Estimated cost |
|---|---|---|---|
| Quick check | 10K / 1K × 1 | 0% | $0.00309 |
| Long analysis | 100K / 4K × 1 | 50% | $0.022 |
| Workday | 20K / 2K × 1,000 | 30% | $5.37 |
More from DeepSeekAll DeepSeek models
| Model | Context | Input / M | Output / M | Intel. |
|---|---|---|---|---|
| DeepSeek V4 Flash 0731 | 1.0M | $0.090 | $0.180 | 49.9 |
| DeepSeek V4 Pro | 1.0M | $0.435 | $0.870 | 44.3 |
| DeepSeek V3.1 Terminus | 164K | $0.270 | $1.00 | 30.4 |
| R1 | 164K | $0.700 | $2.50 | 18.5 |
| DeepSeek V3 0324 | 164K | $0.270 | $1.12 | 15.4 |