All LLM Models
Browse 49 LLM models with VRAM requirements, quantization options, and hardware compatibility.
Model List
DeepSeek V3.1 Terminus
DeepSeek · 684.5B · runs from 412.2 GB at Q4
9.8K
386DeepSeek-V3.1-Terminus is an updated checkpoint of DeepSeek-V3.1, a Mixture-of-Experts chat and reasoning model with roughly 40.1 billion active parameters out of about 684.5 billion total, fine-tuned from DeepSeek-V3.1-Base. This revision keeps the same capabilities as V3.1 while fixing issues reported by users, chiefly reducing mixed Chinese-English text and stray characters in output and further improving the model's Code Agent and Search Agent performance, including gains on BrowseComp, SWE-bench Verified, and Terminal-bench. Its scale requires a multi-GPU server; it is not something that fits on consumer hardware. Context length is 163,840 tokens. It is released under the MIT license, permitting unrestricted commercial and research use. It was published in September 2025, as a refinement of DeepSeek-V3.1 rather than a new base model.
DeepSeek OCR
DeepSeek · 3.3B · runs from 2.8 GB at Q4
DeepSeek OCR is a 3.3-billion-parameter vision-language model from DeepSeek, purpose-built for optical character recognition and document parsing rather than general chat. It pairs a vision encoder with a Mixture-of-Experts decoder that activates roughly 1.1 billion parameters per token, keeping decoding fast while all expert weights still need to fit in memory. Its core idea is compressing a page of text into a much smaller set of image tokens before decoding, and it is small enough to run on a single consumer GPU once quantized. Context length is limited to 8,192 tokens, reflecting its page-oriented use case. It is released under the MIT license, a highly permissive option for commercial use, and was published in October 2025, introducing "optical context compression" to shrink the token count needed for OCR.
DeepSeek V2.5 1210
DeepSeek · 235.7B · runs from 142.9 GB at Q4
DeepSeek-V2.5-1210 is an updated checkpoint of DeepSeek's DeepSeek-V2.5 chat model, a mixture-of-experts model that routes across 160 experts (plus 2 always-on shared experts) with 6 selected per token, for roughly 21.4 billion active parameters out of about 236 billion total. Compared to the original DeepSeek-V2.5, this revision improves math (MATH-500 from 74.8% to 82.8%), coding (LiveCodeBench from 29.2% to 34.38%), and writing and reasoning quality, plus smoother file-upload and webpage-summarization behavior. The card states that full-precision inference needs an eight-GPU server, so local use realistically means a multi-GPU setup even after heavy quantization. Context length is 163,840 tokens. It is released under DeepSeek's custom Model License Agreement, and was published in December 2024, about a year before the DeepSeek-V3 family.
DeepSeek V4 Flash DSpark
DeepSeek · 304.2B · runs from 183.0 GB at Q4
DeepSeek V4 Flash DSpark packs roughly 165 billion parameters into a Flash-designated variant of DeepSeek's newer V4 line, aimed at faster inference than the standard release. It is a chat-oriented model intended for general dialogue and instruction-following tasks rather than a specialized one, continuing DeepSeek's open-weight release pattern. It supports a 1M token context window, well suited to long-document and long-context workloads that need to retain a great deal of information in a single pass. It is released under the MIT license for broad local and commercial use. At this parameter count, DeepSeek V4 Flash DSpark needs multi-GPU or server-class hardware to run locally, even once quantized.
DeepSeek V3.2 Exp
DeepSeek · 685.4B · runs from 412.7 GB at Q4
DeepSeek-V3.2-Exp is an experimental release from DeepSeek that introduces DeepSeek Sparse Attention, a fine-grained sparse attention mechanism meant to improve training and inference efficiency for long-context work. It is a Mixture-of-Experts model at roughly 685 billion total parameters, fine-tuned from a matching V3.2-Exp base checkpoint, and it retains the general reasoning, coding, and agentic tool-use capabilities of DeepSeek's earlier V3.1-Terminus release while testing whether sparse attention can match that quality more efficiently. It supports a 160K token context window and is released under the MIT license. At roughly 685 billion total parameters, even 4-bit quantization needs several hundred gigabytes of memory, so realistic local use means multi-GPU server hardware; most users will rely on a hosted endpoint instead.
DeepSeek v2 Lite
DeepSeek · 15.7B · runs from 10.2 GB at Q4
DeepSeek V2 Lite is a compact mixture-of-experts model with 15.7 billion total parameters, designed to deliver a strong quality-to-compute ratio for general chat and instruction following. It uses the same innovative MLA (Multi-Head Latent Attention) architecture as the larger V2, which reduces memory requirements during inference. With its modest parameter count, V2 Lite runs comfortably on a single consumer GPU, making it accessible to users who want to try DeepSeek's MoE approach without needing specialized hardware. It handles everyday conversational tasks, summarization, and light analysis well, offering a practical entry point into the DeepSeek model family.
DeepSeek V3.2 Speciale
DeepSeek · 685.4B · runs from 412.7 GB at Q4
DeepSeek-V3.2-Speciale is DeepSeek's high-compute reasoning variant of DeepSeek-V3.2, a mixture-of-experts model with roughly 41 billion active parameters out of about 685 billion total, fine-tuned from DeepSeek-V3.2-Exp-Base. It uses DeepSeek Sparse Attention (DSA), an efficient attention mechanism aimed at long-context scenarios, together with heavy reinforcement-learning post-training; the card reports gold-medal-level performance at the 2025 International Mathematical Olympiad and International Olympiad in Informatics, and claims it surpasses GPT-5 with reasoning on par with Gemini 3.0 Pro. Unlike the standard DeepSeek-V3.2 checkpoint, Speciale is dedicated purely to deep reasoning and does not support tool-calling. At this scale it requires a multi-GPU server cluster to run even quantized. Context length is 163,840 tokens. It is released under the MIT license, permitting unrestricted commercial and research use, and was published in November 2025.
DeepSeek v2
DeepSeek · 235.7B · runs from 142.9 GB at Q4
DeepSeek-V2 is DeepSeek's base pretrained Mixture-of-Experts language model, not instruction-tuned, comprising about 235.7 billion total parameters with roughly 21.4 billion activated per token. It was pretrained on 8.1 trillion tokens and built for economical training and efficient inference, cutting training costs and KV-cache size sharply compared with DeepSeek's earlier dense 67B model while boosting generation throughput. Instruction-tuned SFT and RL chat variants were released separately as DeepSeek-V2-Chat. Despite the large total parameter count, its MoE design activates only a fraction per token, but it still needs a multi-GPU server to run at full precision. Context length is 163,840 tokens. It is released under a custom DeepSeek Model License that permits free use, including commercial use, subject to acceptable-use restrictions in the license agreement. It was published in April 2024, ahead of the DeepSeek-V2-Chat and later DeepSeek-V3 models.
Deepseek Vl2 Tiny
DeepSeek · 3.4B · runs from 4.2 GB at Q4
DeepSeek-VL2-Tiny is the smallest in DeepSeek's VL2 series of Mixture-of-Experts vision-language models, with about 3.4 billion total parameters but only roughly 1.2 billion activated per token. It handles vision-language tasks — visual question answering, OCR, document/table/chart understanding, and visual grounding — below the larger VL2-Small and full VL2 variants. Only a fraction of parameters activate per token, so decoding stays fast even though every expert must still be loaded into memory. It is compact enough to run on a single consumer GPU once quantized. Context length is limited to 4,096 tokens. It is released under DeepSeek's own model license, a custom permissive license with an acceptable-use policy rather than a fully open license like MIT or Apache 2.0. Published in December 2024, it pairs a small DeepSeekMoE-3B backbone with a dynamic image-tiling vision encoder.
Deepseek Moe 16B Base
DeepSeek · 16.4B · runs from 11.1 GB at Q4
Deepseek Moe 16B Base is a 16.4B-parameter open language model from DeepSeek in the DeepSeek family. It supports a context window of up to 4,096 tokens. See its VRAM requirements by quantization and which GPUs and Macs can run it locally below.
Deepseek Coder 1.3B Instruct
DeepSeek · 1.3B · runs from 4.3 GB at Q4
DeepSeek Coder 1.3B Instruct is an ultra-compact code model designed for environments where hardware resources are extremely limited. Despite having just 1.3 billion parameters, it can handle basic code completion, simple generation tasks, and code Q&A across common programming languages. This is one of the smallest viable code models available, capable of running on integrated graphics or very low-end dedicated GPUs. It is well suited for edge deployment, embedded development environments, or as a fast local autocomplete engine where response speed matters more than handling complex multi-file reasoning tasks.
Deepseek Llm 7B Base
DeepSeek · 7B · runs from 6.5 GB at Q4
Deepseek Llm 7B Base is a 7B-parameter open language model from DeepSeek in the DeepSeek family. It supports a context window of up to 4,096 tokens. See its VRAM requirements by quantization and which GPUs and Macs can run it locally below.
DeepSeek v2 Chat
DeepSeek · 235.7B · runs from 142.9 GB at Q4
DeepSeek v2 Chat is a 235.7B-parameter open language model from DeepSeek in the DeepSeek V2 family. It supports a context window of up to 163,840 tokens. See its VRAM requirements by quantization and which GPUs and Macs can run it locally below.
Deepseek Moe 16B Chat
DeepSeek · 16.4B · runs from 11.1 GB at Q4
Deepseek Moe 16B Chat is a 16.4B-parameter open language model from DeepSeek in the DeepSeek family. It supports a context window of up to 4,096 tokens. See its VRAM requirements by quantization and which GPUs and Macs can run it locally below.
Deepseek Llm 7B Chat
DeepSeek · 7B · runs from 6.5 GB at Q4
Deepseek Llm 7B Chat is a 7B-parameter open language model from DeepSeek in the DeepSeek family. It supports a context window of up to 4,096 tokens. See its VRAM requirements by quantization and which GPUs and Macs can run it locally below.
Deepseek Coder 1.3B Base
DeepSeek · 1.3B · runs from 4.3 GB at Q4
Deepseek Coder 1.3B Base is a 1.3B-parameter open language model from DeepSeek in the DeepSeek Coder family. It supports a context window of up to 16,384 tokens. See its VRAM requirements by quantization and which GPUs and Macs can run it locally below.
DeepSeek Coder v2 Lite Base
DeepSeek · 15.7B · runs from 10.2 GB at Q4
DeepSeek Coder v2 Lite Base is a 15.7B-parameter open language model from DeepSeek in the DeepSeek Coder family. It supports a context window of up to 163,840 tokens. See its VRAM requirements by quantization and which GPUs and Macs can run it locally below.
DeepSeek R1 Zero
DeepSeek · 684.5B · runs from 412.2 GB at Q4
DeepSeek R1 Zero is a 684.5B-parameter open language model from DeepSeek in the DeepSeek R1 family. It supports a context window of up to 163,840 tokens. See its VRAM requirements by quantization and which GPUs and Macs can run it locally below.
DeepSeek V4 Pro DSpark
DeepSeek · 1650.5B · runs from 990.8 GB at Q4
DeepSeek-V4-Pro-DSpark is not a distinct model but the DeepSeek-V4-Pro checkpoint — a 1.6-trillion-parameter mixture-of-experts model — packaged with an additional DSpark speculative-decoding module bolted on for faster inference; DeepSeek's own card lists 49 billion parameters active per token for the underlying DeepSeek-V4-Pro checkpoint. The V4 series introduces a hybrid attention design combining Compressed Sparse Attention and Heavily Compressed Attention for long-context efficiency, Manifold-Constrained Hyper-Connections for training stability, and the Muon optimizer, and was pretrained on more than 32 trillion tokens. It sits alongside a smaller 284B-total/13B-active V4-Flash sibling that shares the same architecture. Given its enormous total size, it requires a multi-GPU server even heavily quantized. Context length is 1,048,576 tokens (1M). It is released under the MIT license, permitting unrestricted commercial and research use, and was published in June 2026.
Understanding LLM VRAM Requirements
How much VRAM you need depends on the model size and quantization level. Quantization reduces the precision of model weights, trading small quality losses for significantly lower VRAM usage. For example, a 7B parameter model needs ~14 GB at FP16 but only ~4 GB at Q4_K_M quantization.