# DeepSeek-V3-Base

[Canonical model page](https://headlessthrones.com/models/deepseek-deepseek-v3-base) · [Model library](https://headlessthrones.com/models)

Public catalog reference from Headless Thrones. Exact catalog ID: `deepseek-deepseek-v3-base`. Catalog source-check date: 2026-10-23.

This brief contains the public catalog answers and any catalog guide, not private drafts, historical editorial revisions or independently measured rankings. The canonical HTML page may also display separately published editorial.

Current catalog answers

## What to know about DeepSeek-V3-Base

AI-compiled from linked model documentation. This is a catalog reference, not an independent performance review. Catalog checked October 23, 2026.

[Exact model](https://headlessthrones.com/models/deepseek-deepseek-v3-base#model-answer-identity) · [Access](https://headlessthrones.com/models/deepseek-deepseek-v3-base#model-answer-access) · [Token limits](https://headlessthrones.com/models/deepseek-deepseek-v3-base#model-answer-context) · [Inputs & outputs](https://headlessthrones.com/models/deepseek-deepseek-v3-base#model-answer-modalities) · [Model details](https://headlessthrones.com/models/deepseek-deepseek-v3-base#model-answer-research) · [Weight details](https://headlessthrones.com/models/deepseek-deepseek-v3-base#model-answer-repository)

### Which DeepSeek-V3-Base does this page describe?

DeepSeek-V3-Base is a model record from **DeepSeek** in the DeepSeek family. [Source](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base)

The recorded model identifier or checkpoint is `DeepSeek-V3-Base`. Use that exact identifier when checking documentation or a provider listing. [Source](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base)

### How can you access DeepSeek-V3-Base?

The catalog documents **Open weights** access  for DeepSeek-V3-Base. [Source](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base)

As of the catalog check, this version was listed as available. [Source](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base)

Open weights means the model weights are distributed for use under their stated terms. It does not establish a hosted API, an open-source license, or a hardware requirement.

### What are DeepSeek-V3-Base's context and output limits?

The publisher reports the context window as **128K tokens**. [Source](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/resolve/afb92e1fa402c2be2a9eb085312bb02e0384d6c7/README.md) We retain the publisher's notation and qualifiers; no exact token count or hosted serving allowance is inferred from it.

A separate maximum-output value is not established in this record. The context or input figure should not be used as an output allowance.

Context, input and output limits describe different constraints. Do not add the figures together to infer a larger window. Check the endpoint's accounting for conversation history, system instructions, tools, media and generated tokens; a documented maximum is not a guarantee of answer quality or a measured usable capacity.

### What can DeepSeek-V3-Base receive and generate?

Documented inputs: **text**. [Source](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/resolve/afb92e1fa402c2be2a9eb085312bb02e0384d6c7/README.md)

Documented outputs: **text**. [Source](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/resolve/afb92e1fa402c2be2a9eb085312bb02e0384d6c7/README.md)

Input support and output generation are separate capabilities. Reading an image does not establish image generation; a text output listing does not establish speech output. File sizes, formats and media limits depend on the documented implementation.

### What distinguishes DeepSeek-V3-Base?

These details come from the linked publisher material for this model. Reported capabilities and training descriptions are publisher statements, not independent test results.

**Nominal expert parameters**: 671 billion main-model parameters; 37 billion activated per token. [Source](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/resolve/afb92e1fa402c2be2a9eb085312bb02e0384d6c7/README.md)

**Checkpoint accounting**: The downloaded checkpoint includes an additional 14B multi-token prediction module beyond the 671B main model. [Source](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/resolve/afb92e1fa402c2be2a9eb085312bb02e0384d6c7/README.md)

### What is documented about the DeepSeek-V3-Base checkpoint?

This evidence refers to `deepseek-ai/DeepSeek-V3-Base` at recorded revision `afb92e1fa402c2be2a9eb085312bb02e0384d6c7`, checked October 23, 2026. [Repository metadata](https://huggingface.co/api/models/deepseek-ai/DeepSeek-V3-Base)

Safetensors metadata records **684,489,845,504 stored parameters**. This is a stored-weight count, not necessarily the number activated per token, a download size or a GPU-memory requirement. [Repository metadata](https://huggingface.co/api/models/deepseek-ai/DeepSeek-V3-Base)

Recorded architecture class: `DeepseekV3ForCausalLM`. [Repository metadata](https://huggingface.co/api/models/deepseek-ai/DeepSeek-V3-Base)

Repository access gate: `false`. This records the repository's gate setting when checked; licensing and use terms still apply. [Repository metadata](https://huggingface.co/api/models/deepseek-ai/DeepSeek-V3-Base)

#### What the configuration file says

The pinned `config.json` contains these values. [Pinned configuration](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/resolve/afb92e1fa402c2be2a9eb085312bb02e0384d6c7/config.json)

`max_position_embeddings`: 163,840

`rope_scaling`: {"beta\_fast":32,"beta\_slow":1,"factor":40,"mscale":1,"mscale\_all\_dim":1,"original\_max\_position\_embeddings":4096,"type":"yarn"}

`torch_dtype`: bfloat16

`num_attention_heads`: 128

`num_key_value_heads`: 128

`num_hidden_layers`: 61

`hidden_size`: 7,168

Configuration values describe the checkpoint and its implementation. In particular, max\_position\_embeddings, sliding\_window and rope\_scaling do not independently establish an API context limit, tested usable length or hardware requirement. Runtime, quantization, cache, batch size and input length affect deployment behavior.

Read the retrieved model card for setup and qualifications. [Model card](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/resolve/afb92e1fa402c2be2a9eb085312bb02e0384d6c7/README.md)

**Catalog qualifications:** Official publisher weight listing; availability refers to this distribution, not a guaranteed hosted API. Base, SFT and research variants are identified by their exact published names. [Source 1](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base) · [Source 2](https://huggingface.co/api/models/deepseek-ai/DeepSeek-V3-Base) · [Source 3](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/resolve/afb92e1fa402c2be2a9eb085312bb02e0384d6c7/README.md) · [Source 4](https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/resolve/afb92e1fa402c2be2a9eb085312bb02e0384d6c7/config.json)

**Evidence still missing here:** knowledge and training-data cutoffs and tested hardware requirements. Missing documentation is not evidence that a feature is unavailable.

Headless Thrones has not independently benchmarked this model for this reference. Evaluate it on representative work before choosing it; these facts do not rank its accuracy, reliability, speed or suitability against another model.

Keep the exact version in view

## Other DeepSeek models from DeepSeek.

These are separate records in the same publisher family. A shared family name does not establish equal capabilities or interchangeable licenses.

### [deepseek-coder-1.3b-base](https://headlessthrones.com/models/deepseek-deepseek-coder-1-3b-base)

Open weights

### [deepseek-coder-1.3b-instruct](https://headlessthrones.com/models/deepseek-deepseek-coder-1-3b-instruct)

Open weights

### [deepseek-coder-33b-base](https://headlessthrones.com/models/deepseek-deepseek-coder-33b-base)

Open weights

### [deepseek-coder-33b-instruct](https://headlessthrones.com/models/deepseek-deepseek-coder-33b-instruct)

Open weights

### [deepseek-coder-5.7bmqa-base](https://headlessthrones.com/models/deepseek-deepseek-coder-5-7bmqa-base)

Open weights

### [deepseek-coder-6.7b-base](https://headlessthrones.com/models/deepseek-deepseek-coder-6-7b-base)

Open weights