Skip to main content
Rase AI Intelligence Newsletter

Models & assistants Model or AI service

DeepSeek-V3 released base and chat model weights

DeepSeek published V3 as a downloadable language model with base and chat versions. It uses a mixture of experts, which activates part of the model for each token. [2][3]

Model access
Base and chat model weights are downloadable. DeepSeek’s own hosted API moved to later versions after launch. [1][2]
License
DeepSeek’s Model License for Base and Chat weights. MIT for repository code. [2]

What the model does

DeepSeek released the original V3 in December 2024 and moved its hosted chat API to V3 at launch. The team published checkpoint files and documentation for running the model. [1][2][3]

Why it matters

Downloading an exact checkpoint lets a team repeat evaluations even when a hosted model changes. V3’s design also shows why total model size and the amount used for each token are different measures. [1][2]

Who can use it

Researchers and hosting teams can access the base and chat weights. Running the original model requires infrastructure able to handle the full checkpoint. [2]

Key dates

  1. Original DeepSeek-V3 released [1][3]

Ways to use it

For a reproducible assessment, record the exact V3 checkpoint and test it on tasks such as text extraction or language generation. Hosted API performance can change when the provider updates the model behind an alias. [1][2]

Model limitations

The main model has 671B parameters and activates 37B per token. The full set of weights still needs storage and memory, and model use is governed by a separate license from the repository code. [2]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source DeepSeek’s official API release history
  2. Primary source DeepSeek-V3 official repository, model downloads and license
  3. Primary source 🚀 Introducing DeepSeek-V3 | DeepSeek API Docs

Related profiles

Models & assistants

GPT-6.2 Astra and Sol

OpenAI

Astra targets demanding projects. Sol targets everyday tasks. Both support text and image input in the API.

High confidenceEvidence checked

Models & assistants

MAI Code 2 and Code 2 Flash

Microsoft AI

Code 2 targets complex software work. Flash targets frequent edits, explanations and routine fixes.

High confidenceEvidence checked

Models & assistants

MAI Realtime 1

Microsoft AI

Users can try a MAI Labs demo. Developers can register for a limited Foundry preview.

High confidenceEvidence checked