Skip to content
TrustList

TrustList summary

A Mixture-of-Experts language model with 671B total parameters and 37B activated for each token, DeepSeek-V3 is published openly by DeepSeek in a GitHub repository under an MIT licence. Context length is 128K, and the repository offers the base model and the chat model for download from Hugging Face.

Architecture choices

DeepSeek says V3 uses Multi-head Latent Attention and the DeepSeekMoE design, both carried over from DeepSeek-V2, for efficient inference and training. It adds a load-balancing strategy that avoids an auxiliary loss, which it says limits the performance cost usually paid for balancing, and a multi-token prediction training objective that can also serve speculative decoding to speed up inference.

Training

The model was pre-trained on 14.8 trillion tokens, followed by supervised fine-tuning and reinforcement learning. DeepSeek reports an FP8 mixed-precision framework, first validated at this scale, and cross-node work that nearly overlaps computation with communication. Full training is reported at 2.788M H800 GPU hours, and the team notes no irrecoverable loss spikes or rollbacks.

Reasoning distilled from R1

For post-training, DeepSeek describes distilling the long chain-of-thought behaviour of a DeepSeek-R1 model into V3, borrowing its verification and reflection patterns while keeping control of output style and length. The repository also covers a chat website, an API platform and instructions for running the model locally, and it claims results that beat other open-source models and match leading closed ones.

Summary written by TrustList from the business's own published information, Nov 2026.

Benchmarks & AI stats

BenchmarkOfficialCommunity avg
No benchmark scores yet. Be the first to add one.

“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).

Sign in to add a benchmark score for this model.

DeepSeek-V3 launches

Request a demo or quote from DeepSeek-V3

Protected by reCAPTCHA - Google Privacy Policy and Terms apply.

By sending, you agree we may share your request and contact details with the provider once you confirm your email.

Reviews

Write the first review of DeepSeek-V3

Used it? Your experience helps other buyers decide.

Write a review

Questions & answers

No questions yet. Be the first to ask about DeepSeek-V3.

DeepSeek-V3 alternatives

Similar listings buyers compare against this one.

Open-source 2B decision model from AWS's Strands Agents project that picks options and returns scores with a reliability figure

Black Forest Labs' image generation and editing model with bounding-box layouts, multiple reference images and 4K output

Microsoft AI's text-to-speech model for keeping one voice across 23 languages, with voice cloning from short audio

Open-weight translation model from Bilibili's Index team, covering 150 languages and following glossary and style instructions

Google's open multimodal embedding model that maps text, images, audio and video into one vector space, under Apache 2.0

Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems

Microsoft AI's real-time speech-to-text model for voice agents, transcribing 60 languages with continuous language detection

Anthropic's smallest and fastest Claude 5.5 model, available on its own API and three cloud platforms