DeepSeek V3.2 Exp: Self-Hosting & Deployment Guide
DeepSeek V3.2 Exp is an experimental iteration of DeepSeek’s large language model series, focused on testing advanced reasoning and generation capabilities before they are incorporated into stable releases.
What is DeepSeek V3.2 Exp?
DeepSeek V3.2 Exp is an experimental large language model developed by DeepSeek to explore improvements in language understanding, reasoning, and response quality. It is primarily used for research, prototyping, and evaluating new techniques in dialogue, content generation, and task automation. It can also support developers and researchers in benchmarking, ablation studies, and explorations of novel prompting or fine-tuning strategies. It belongs to the broader DeepSeek model family, extending the capabilities and design ideas of earlier DeepSeek V-series models.
How to run DeepSeek V3.2 Exp yourself
Open weights (mit), 685.4B parameters. Below: how much memory the weights need at each quantization, which hardware fits them, and the commands to serve it.
DeepSeek V3.2 Exp VRAM requirements
| Precision / quantization | Weights | Figure | Source |
|---|---|---|---|
| BF16 (full precision) | 1370.8 GB | Estimate | 685.4B params × 2 bytes |
| AWQ 4-bit | 399.8 GB | File size | QuantTrio/DeepSeek-V3.2-Exp-AWQ |
| GGUF Q8_0 | 713.3 GB | File size | sszymczyk/DeepSeek-V3.2-Exp-nolight-GGUF |
| GGUF Q4_K_M | 404.5 GB | File size | sszymczyk/DeepSeek-V3.2-Exp-nolight-GGUF |
File size = actual download size on Hugging Face. Estimate = parameter count × bytes per weight. Weights only — the KV cache for long contexts needs extra memory; the fit tables below add 15% headroom as a rule of thumb.
GPU requirements: deploy DeepSeek V3.2 Exp on your hardware
| Hardware | BF16 (full precision) | AWQ 4-bit | GGUF Q8_0 | GGUF Q4_K_M |
|---|---|---|---|---|
| NVIDIA H200141 GB | — | 4× | 6× | 4× |
| NVIDIA H10080 GB | — | 6× | — | 6× |
| NVIDIA A100 80GB80 GB | — | 6× | — | 6× |
| NVIDIA L40S48 GB | — | — | — | — |
GPUs needed per variant.
| Hardware | GGUF Q8_0 | GGUF Q4_K_M |
|---|---|---|
| RTX 509032 GB | — | — |
| RTX 409024 GB | — | — |
| RTX 309024 GB | — | — |
GGUF via llama.cpp; more cards or CPU offload for larger files.
| Hardware | GGUF Q8_0 | GGUF Q4_K_M |
|---|---|---|
| Mac, 512 GBM3 Ultra | — | — |
| Mac, 192 GBM2 Ultra | — | — |
| Mac, 128 GBM4 Max / M3 Max | — | — |
| Mac, 64 GBM4 Pro / Max | — | — |
Assumes ~75% of unified memory is usable by the GPU.
| Hardware | BF16 (full precision) | AWQ 4-bit | GGUF Q8_0 | GGUF Q4_K_M |
|---|---|---|---|---|
| AWS p5.48xlarge8× H100 · 640 GB | — | Fits | — | Fits |
| AWS p4de.24xlarge8× A100 80GB · 640 GB | — | Fits | — | Fits |
| AWS g6e.12xlarge4× L40S · 192 GB | — | — | — | — |
| Google Cloud a3-highgpu-8g8× H100 · 640 GB | — | Fits | — | Fits |
| Azure ND H100 v58× H100 · 640 GB | — | Fits | — | Fits |
Whole-instance GPU memory; public instance specs.
Serve it yourself
pip install -U vllm
vllm serve deepseek-ai/DeepSeek-V3.2-Exppip install 'sglang[all]'
python -m sglang.launch_server --model-path deepseek-ai/DeepSeek-V3.2-Expllama-server -hf sszymczyk/DeepSeek-V3.2-Exp-nolight-GGUF:Q4_K_M \
--ctx-size 32768 --n-gpu-layers 99Commands use the official repositories above; vLLM and SGLang expose an OpenAI-compatible endpoint.
Not on LLM.API yet
We do not route this model through our API at the moment. Browse the models you can call today — most workloads have a close match already live.
Providers
Different companies host the same model. OpenRouter routes your request to one of them based on the routing mode you pick — Balanced (price + speed), Nitro (fastest), or Exacto (highest tool-calling accuracy).
List price by provider
InputOutputProvider list prices; the LLM.API discount applies on top.
| Provider | Input | Output | Cache read /M | Latency | Throughput | Uptime |
|---|---|---|---|---|---|---|
| in $0.0003; out $0.0004 per 1M tokens | 164K tokens | tools, streaming, reasoning, JSON, structured |
List prices from third-party hosts in the OpenRouter public catalogue (deepseek/deepseek-v3.2-exp); LLM API does not serve this model and applies no discount. Updated weekly. Last updated 11 Oct 2026.
Try this model
Test DeepSeek V3.2 Exp right here — free to start.
Suggestions for your first prompt
Not available on LLM API yet
We do not route this model through our API at the moment, so there is no endpoint or code snippet for it yet. Browse the models you can call today — most workloads have a close match already live.
Browse available modelsDeepSeek V3.2 Exp: Self-Hosting & Deployment Guide in production
What the model is good at.
-
Conversational Chat
Engages in multi-turn dialogue, follows instructions, and maintains context to answer questions or assist with complex tasks.
-
Code Reasoning
Understands and generates code snippets, explains programming concepts, and helps debug or refactor code across common languages.
-
Text Translation
Translates text between multiple languages while aiming to preserve original meaning, tone, and key domain terminology.
-
Document OCR
Extracts machine-readable text from images or scanned documents, supporting downstream search, analysis, and transformation workflows.
-
Image Understanding
Interprets images by describing contents, detecting objects, and providing contextual insights about scenes or visual elements.
What you get when you call DeepSeek V3.2 Exp: Self-Hosting & Deployment Guide through LLM.API.
-
Intelligent Model Routing
Dynamically route each request to the optimal model across providers based on latency, price, and quality—without changing your integration or redeploying code.
One API, many models -
Cost-Aware Orchestration
Automatically balance premium and budget models with per-call controls, caps, and policies so you can ship fast while keeping AI spend predictable and optimized.
Control spend by design -
Resilient Fallback Flows
Define cascading provider and model fallbacks so timeouts, quota limits, or regional outages transparently fail over—maintaining uptime without custom retry logic.
Fail soft, not hard -
End-to-End Observability
Trace every request across providers with logs, metrics, and structured payloads to debug prompts, compare models, and tune performance from a single dashboard.
See every token -
Task-Level Abstractions
Call high-level tasks like chat, generate, extract, or rank instead of vendor-specific APIs, so you can swap models without rewriting business logic.
Code to tasks, not vendors -
High-Throughput Batch Jobs
Run large-scale inference workloads as managed batches with concurrency, retries, and progress tracking built in—perfect for backfills, evaluations, and data processing.
Crush your backlogs
When to Use — When NOT to Use
Use it if...
- You need a cost-effective general-purpose model for everyday coding, writing, and chatting.
- You need rapid iteration on prototypes where slightly lower reliability is acceptable for speed.
- You need an assistant for routine code refactoring, small bug fixes, and style improvements.
- Your use case involves generating short-form content like emails, summaries, or marketing blurbs.
- Your use case involves lightweight data cleaning, simple transformations, and basic text classification.
- You need an inexpensive model for A/B testing prompts before moving to premium models.
Avoid if...
- You need guaranteed state-of-the-art reasoning performance comparable to the very strongest frontier models.
- Your workload requires strict enterprise compliance certifications and audited security guarantees from major vendors.
- You need highly reliable execution of complex multi-step tools, agents, or autonomous workflows.
- Your workload requires extremely long-context processing of hundreds of pages with minimal hallucinations.
- You need robust safety, red-teaming, and policy controls proven in large-scale regulated deployments.
- Your workload requires tight integration with a specific commercial ecosystem or managed cloud platform.
COMMUNITY
What developers say about DeepSeek V3.2 Exp
Summarised from public developer discussions and write-ups. Opinions are the sources’, not LLM.API’s.
- Developers on Hacker News noted that the pace of model scaling significantly correlates with rapid price declines, suggesting that optimization and supply chain factors are driving costs down faster than Moore's Law.
- Early users and experts report that the model's new Sparse Attention mechanism facilitates 2–3× faster long-text processing and a 30–40% reduction in memory usage compared to previous versions.
- Community leads on Hugging Face emphasize that the model makes high-end AI more accessible to smaller companies by cutting running costs in half without a noticeable drop in performance quality.
SOURCES
Frequently Asked Questions
-
What is DeepSeek V3.2 Exp?
DeepSeek V3.2 Exp is an experimental DeepSeek large language model focused on fast, low-cost text generation for developer workloads.
-
What is DeepSeek V3.2 Exp best suited for?
DeepSeek V3.2 Exp is best suited for general coding assistance, tool-using agents, and high-volume chat or completion workloads where cost efficiency matters.
-
What is the context window of DeepSeek V3.2 Exp via LLM.API?
DeepSeek V3.2 Exp supports a 32K token context window when accessed through LLM.API.
-
What modalities does DeepSeek V3.2 Exp support?
DeepSeek V3.2 Exp currently supports text-in, text-out interactions only through LLM.API.
-
How is DeepSeek V3.2 Exp priced on LLM.API?
DeepSeek V3.2 Exp uses LLM.API’s unified per-token pricing; check your LLM.API dashboard for the latest input and output token rates.
-
How fast is DeepSeek V3.2 Exp in terms of latency and throughput?
DeepSeek V3.2 Exp is optimized for low latency and high throughput, making it suitable for real-time applications and parallel request loads.
-
How do I call DeepSeek V3.2 Exp through the LLM.API?
Use the LLM.API chat or completion endpoint with the model identifier "deepseek-v3.2-exp" and your standard LLM.API authentication header.
-
How does DeepSeek V3.2 Exp compare to similar DeepSeek or OpenAI models?
DeepSeek V3.2 Exp generally trades some reasoning depth for higher speed and lower cost compared with frontier flagship models of similar size.
-
Does DeepSeek V3.2 Exp support function calling or tools via LLM.API?
Yes, DeepSeek V3.2 Exp supports structured tool or function calling when you pass a tools schema to the LLM.API chat endpoint.
-
What are the main limitations of DeepSeek V3.2 Exp?
DeepSeek V3.2 Exp may hallucinate facts, lacks up-to-the-minute knowledge, and should not be used as the sole source for critical decisions.
COMPARE
Competitive Models
-
DeepSeek V4 Flash
Up to 30%
DeepSeek V4 Flash is an efficiency-optimized Mixture-of-Experts large language model from DeepSeek, featuring a 1M-token context window and fast inference for high-throughput applications.
-
DeepSeek V4 Flash (free)
Up to 30%
DeepSeek V4 Flash (free) is an open-source, efficiency-optimized Mixture-of-Experts language model from DeepSeek, offering a 1M-token context window with only 13B parameters activated per token out of 284B total. It is designed to deliver fast, cost-effective long-context reasoning, coding, and agentic workflows.
-
DeepSeek V3.2
DeepSeek V3.2 is a large open-source Mixture-of-Experts language model from DeepSeek that emphasizes high reasoning performance and efficient long‑context inference. It is notable for its DeepSeek Sparse Attention and multi-latent attention mechanisms, which significantly cut compute and memory costs for long sequences.
Get one key to every model
Swap your API key. Keep your code.