# DeepSeek models on W&B Inference

## DeepSeek V3-0324 inference overview

#### Price per 1M tokens

$1.14 (input)

$2.75 (output)

#### Parameters

37B

680B

#### Context window

161K

#### Release date

Mar 2025

### DeepSeek V3-0324 inference details

DeepSeek V3-0324 is a powerful Mixture-of-Experts model designed for demanding language tasks, including comprehensive information extraction, content summarization, document analysis, and handling complex, structured textual data. It excels in balancing depth and accuracy in high-complexity scenarios.

**Created by**: DeepSeek  
**License**: mit  
**🤗 model card**: [DeepSeek-V3-0324](https://huggingface.co/deepseek-ai/DeepSeek-V3-0324)

```python

```

## DeepSeek R1-0528 inference overview

#### Price per 1M tokens

$1.35 (input)

$5.40 (output)

#### Parameters

37B

680B

#### Context window

161K

#### Release date

May 2025

### DeepSeek R1-0528 inference details

DeepSeek R1-0528 specializes in tasks requiring detailed reasoning, including mathematics, programming, and logical problem-solving. It’s particularly effective in scenarios like complex coding tasks, planning workflows, and analyzing structured documents with enhanced accuracy and reduced hallucinations.

**Created by**: DeepSeek  
**License**: mit  
**🤗 model card**: [DeepSeek-R1-0528](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528)

```python

```

## Deepseek V3.1 on W&B Inference

#### Price per 1M tokens

$0.55 (input)

$1.65 (output)

#### Parameters

681B

#### Context window

128K

#### Release date

Aug 2025

### Deepseek V3.1 inference details

DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes via prompt templates. It extends the DeepSeek-V3 base with a two-phase long-context training process, reaching up to 128K tokens, and uses FP8 microscaling for efficient inference. The model improves tool use, code generation, and reasoning efficiency, achieving performance comparable to DeepSeek R1-0528 on difficult benchmarks while responding more quickly. It supports structured tool calling, code agents, and search agents, making it suitable for research, coding, and agentic workflows. It succeeds the DeepSeek V3-0324 model and performs well on a variety of tasks.

**Created by**: DeepSeek  
**License**: mit  
**🤗 model card**: [DeepSeek-V3.1](https://huggingface.co/deepseek-ai/DeepSeek-V3.1)

```python

```

## DeepSeek V3 resources

### A primer on building successful AI agents

[Read the whitepaper](https://wandb.ai/site/resources/whitepapers/building-successful-ai-agents/)

### W&B Inference powered by CoreWeave guide

[Get started](https://wandb.ai/wandb_fc/product-announcements-fc/reports/Introducing-W-B-Inference-powered-by-CoreWeave--VmlldzoxMzI1NjU0MQ)

### AI engineering course: Agents

[Take the free course](https://wandb.ai/site/courses/agents/)