DeepSeek-V3.2
Official · FP8 · safetensors
Runtime certified: Transformers
| Precision / quantization | Format | Bits | Publisher trust | Deployment state |
|---|---|---|---|---|
| FP8 | safetensors | 8 | Official | Runtime certified |
DeepSeek · General · DeepSeek V3.2
DeepSeek V3.2: Specs, GPU requirements, hosting & deployment
DeepSeek V3.2 is a recent General release in the DeepSeek family, certified from its immutable official source revision.
Total parameters671.0B
Active parameters37.0B
Context window163,840 tokens
Release statusRecent · Recent
ArchitectureDeepseekV32ForCausalLM
ModalitiesText
LicenseMIT License
Deployment stateDeployable with certified configuration
Commercial stateCommercial Use Allowed
Release date2025-12-01
Architecture, strengths and use cases
DeepSeek V3.2 is an official self-hostable DeepSeek release in the DeepSeek General track. It uses DeepseekV32ForCausalLM, carries 671B / 37B MoE, supports a verified context window of 163,840 tokens, and exposes only source-certified checkpoints and runtime combinations. Its release, license and deployment facts are tied to the official repository revision rather than inferred from the family name.
DeepSeek V3.2 uses DeepseekV32ForCausalLM with 671B / 37B MoE. The stored architecture profile records 61 layers, hidden width 7168, 128 attention heads and a 163,840 token context ceiling where the official configuration publishes those fields.
Verified DeepseekV32ForCausalLM architecture
163,840 token certified context
Official FP8 checkpoint coverage
Private LLM inference
Agentic workflows
Long-context document processing
Certified checkpoints and quantizations
Only exact, source-certified variants attached to this release are shown.
Recommended hardware
Use only the listed transformers runtime configuration with the exact checkpoint precision. The recommendation engine checks weight memory, KV cache, runtime overhead, safety headroom, GPU architecture and topology before ranking rentable and reference hardware.
Estimated weight memory643.66 GB
KV cache6.67 GB
Runtime overhead173.7 GB
Total with safety margin1021.81 GB
Technically compatible reference sizing
No current FusionPointAI inventory passes every runtime, architecture, topology and memory gate for this exact checkpoint. Use the estimator for global compatible hardware.
License and self-hosting considerations
The full 671B resident weight set must fit across the selected GPUs. Context and concurrency add KV-cache memory; MoE active parameters do not replace resident-weight memory.
DeepSeek V3.2 is published under MIT License. The release-scoped official license link and verification date are retained with the catalogue record.
Commercial Use Allowed. The official license permits commercial use; its obligations still apply.
Alternatives and sibling variants
DeepSeek V3.2 Speciale, DeepSeek V4 Pro 0813, DeepSeek V4 Flash 0731
Release timeline
deepseek/deepseek/deepseek-v3-2-standard
Source details
Source revision: 0d60c440c323cfb1fd82507fa6d120a30f4c1e7c74d17d2f414d713af6c27139
Frequently asked questions
How much VRAM does DeepSeek V3.2 require?
The exact estimate depends on checkpoint precision, context and concurrency; the page recommendation uses the certified memory engine and compatible runtime configurations.
Which runtimes support DeepSeek V3.2?
Certified upstream runtime evidence is available for transformers.
Can DeepSeek V3.2 be used commercially?
The stored commercial classification is Commercial Use Allowed. Consult the official MIT License terms for the final obligations.
Official references
Last verified: 2026-08-26