DeepSeek-V3.2 vs Hy4-preview
Specs, VRAM requirements and download trends — updated 25 October 2026.
deepseek-ai · A · Open weights
Params
685.4 B
Context
163,840
30d
5.6 M
mit · commercial OK
tencent · B · Open weights
Params
780.0 B
Context
1,048,576
30d
18 K
apache-2.0 · commercial OK
Specification comparison
Differences are highlighted; identical values are muted.
| Specification | DeepSeek-V3.2 | Hy4-preview |
|---|---|---|
| Deployment | Open weights · also via API | Open weights · also via API |
| Parameters | 685.4 B | 780.0 B |
| Architecture | DeepseekV32ForCausalLM | HYV4ForCausalLM |
| Context length | 163,840 | 1,048,576 |
| Licence | mit | apache-2.0 |
| Commercial use | Allowed | Allowed |
| Languages | — | — |
| Downloads 30d | 5,571,243 | 18,369 |
| Downloads all time | 30.7 M | 39 K |
| Quantizations on the Hub | SAFETENSORS | SAFETENSORS |
| Gated | No | No |
| First seen on the Hub | 2025-12-01 | 2026-08-27 |
| API input price / 1M tokens | $0.28 | $0.83 |
| API output price / 1M tokens | $0.42 | $2.50 |
| Est. cost per 1M input + 1M output tokens | $0.70 | $3.34 |
| Cached input / 1M tokens | $0.028 | $0.042 |
| AA Intelligence Index | 21.5 | — |
| AA Coding Index | 44.2 | — |
| AA Agentic Index | — | — |
| Max output tokens | 65,536 | 64,000 |
| API providers | 11 | 5 |
| Tool calling | Yes | Yes |
| Reasoning mode | Yes | Yes |
API cost
Estimated cost of 1M input + 1M output tokens via OpenRouter
DeepSeek-V3.2
$0.70
Hy4-preview
$3.34
DeepSeek-V3.2 is 4.8× cheaper at list price.
Download trend
Daily snapshots, last 90 days (28 Jul – 25 Oct)
5.6 M18 K
DeepSeek-V3.2
Hy4-preview
VRAM side by side
On RTX 4090 · 24 GB · 8K context unless noted
DeepSeek-V3.2 · fp16 @ 160K ctx
2,815.1 GB / 24 GB
Hy4-preview · fp16 @ 1024K ctx
16,691.7 GB / 24 GB