DeepSeek V3Open Weight

Released

DeepSeek V3 is an AI model released by DeepSeek on Thursday, Dec 26 2024, 16 days after DeepSeek V2.5 (Revised). It is an open-weight model — the trained weights are available to download and run. It is a 671B parameterThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. model with a 128k token context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider.. Benchmark results (shown below) cover SWE-Bench Verified, LiveCodeBench, GPQA Diamond, and MMLU.

Available from

DeepInfraflex$0.256$0.712160Kfp4
DeepInfra—$0.32$0.89160Kfp4
StreamLake—$0.2574$1.0287128K—
USD per 1M tokensChecked November 17, 2026Third-party rates from openrouter.ai, not the lab’s own list price.

Benchmarks

Coding

SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better.
42%
#73 of 87
LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better.
40.5%
#45 of 60

Reasoning & science

GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better.
59.1%
#107 of 125
MMLUGeneral knowledge — A 57-subject multiple-choice exam — history, law, medicine, maths — that was the standard measure of how much a model knows from 2020 until roughly 2024, when frontier scores crowded into the high 80s and labs moved on to harder tests. The scores here were published years apart under different testing setups, so read them as a historical record rather than a like-for-like ranking. Higher is better.
88.5%
#11 of 59

Compare DeepSeek V3 with

DeepSeek V3

About

DeepSeek V3 was released on December 26, 2024 as a mixture-of-experts language-model family with 671 billion main-model parameters and about 37 billion active per token. DeepSeek published the pretrained Base checkpoint and the post-trained chat model together. Base offered a foundation for further adaptation; the chat model added supervised fine-tuning and reinforcement learning for assistant use. All benchmark scores displayed here represent the chat model.

V3 was pretrained on 14.8 trillion tokens. It retained Multi-head Latent Attention from V2 and introduced auxiliary-loss-free expert balancing, FP8 mixed-precision training and a multi-token prediction objective to improve efficiency. The release package included additional prediction-module weights beyond the headline main-model size. This entry groups the simultaneous Base and chat release while keeping their evaluation results separate.

Frequently asked questions

DeepSeek V3 was released by DeepSeek on Thursday, Dec 26 2024.

All DeepSeek releases

30 tracked