DeepSeek-V3.2-Exp
We are excited to announce the official release of DeepSeek-V3.2-Exp, an experimental version of our model. As an intermediate step toward our next-generation architecture, V3.2-Exp builds upon V3.1-Terminus by introducing DeepSeek Sparse Attention—a sparse attention mechanism designed to explore and validate optimizations for training and inference efficiency in long-context scenarios.
Params
685.4 B
Context
163,840
Downloads 30d
170 K
Likes
995
Download history
daily snapshots · 56 days
▲ 0 in the last 30 days (0.0%)
170 K170 K
Aug 23Sep 2Sep 12Sep 21
288 K170 K
Jul 28Aug 15Sep 3Sep 21
Can you run it?
Estimated VRAM at 8K context unless noted. Pick your hardware to see the verdict per quantization.
| File | Quant | Size | Est. VRAM | Verdict on RTX 4090 · 24 GB |
|---|---|---|---|---|
| model.safetensors | f8_e4m3 | 689.5 GB | 861.7 GB | ❌ Won’t fit |
| model.safetensors (bf16, full) | bf16 + 160K ctx | 689.5 GB | 2,815.1 GB | ❌ Won’t fit |
Estimate: file size × 1.1 + KV cache at 8K + 0.5 GB overhead. Not a benchmark — how we calculate this.
Run it
copy-paste, exact tags checked against the Hub$ curl -s https://aimodelscomparison.com/api/v1/models/deepseek-v3-2-exp
{
"hf_id": "deepseek-ai/DeepSeek-V3.2-Exp",
"params_b": 685.40,
"context_length": 163840,
"license": { "id": "mit", "commercial": "yes" },
"downloads_30d": 170167,
"vram_estimates": [
{ "quant": "f8_e4m3", "gb": 861.7 }
],
"updated_at": "2026-07-28T18:06:54Z"
}
Specifications
- Architecture
- DeepseekV32ForCausalLM
- Parameters
- 685.4 B
- Tensor type
- F8_E4M3
- Context length
- 163,840
- Vocabulary
- 129,280
- Layers / heads
- 61 / 128
- Licence
- mit
- First seen on the Hub
- 2025-09-29
- Training datasets
- undisclosed
- Added to our catalog
- 2026-07-28
Compare with any text-generation model