Model catalog

DeepSeek Models

Browse the AI models associated with DeepSeek. Compare current and historical models by family, capabilities, context window, availability and intended use.

55 models tracked
55 Total models
25 Model families
5 Model types
50 Current / accessible
All models

DeepSeek model catalog

DeepSeek logo
DeepSeek-Coder-V2

DeepSeek-Coder-V2

Self-hosted code generation, completion, repository analysis, debugging, code translation, and programming-focused research.

Type Coding
Context 128K
Reasoning 7/10
Speed 5/10
Streaming
Status

Legacy open-weight model series; downloadable checkpoints remain available, while the hosted Coder API line was superseded and merged into DeepSeek-V2.5.

View model →

Local code completion, fill-in-the-middle generation, IDE integrations, repository-level coding, and private or self-hosted inference

Type Coding
Context 131K
Reasoning 6/10
Speed 8/10
Status

Open-weight and downloadable; accessible through the official Hugging Face repository, with no verified provider-published shutdown date

View model →

Local code generation, code completion, debugging, code explanation, repository-scale prompts, and developers needing an open-weight coding model

Type Coding
Context 128K
Reasoning 7/10
Speed 7/10
Fine-tuning
Status

Available as an open-weight downloadable model; older but still accessible

View model →
DeepSeek logo
DeepSeek-LLM

DeepSeek-LLM 7B

Local bilingual text generation, research, experimentation, instruction tuning, and lightweight self-hosted assistants

Type General Purpose
Context 4K
Reasoning 5/10
Speed 6/10
Fine-tuning Streaming
Status

Legacy open-weight model; downloadable and usable through compatible local inference tools

Input No official hosted API price; downloadable weights for self-hosting
Output No official hosted API price; self-hosting and infrastructure costs apply
View model →
DeepSeek logo
DeepSeek-LLM

DeepSeek-LLM 67B

Local deployment, bilingual English-Chinese generation, language-model research, mathematics, coding experiments, and custom fine-tuning workflows

Type General Purpose
Context 4K
Reasoning 6/10
Speed 3/10
Status

Legacy open-weight model; downloadable checkpoints remain available, but it is superseded by newer DeepSeek model families

Input No official hosted API pricing; downloadable weights
Output No official hosted API pricing; downloadable weights
View model →
DeepSeek logo
DeepSeek-Math

DeepSeek-Math-V2

Advanced mathematical reasoning, natural-language theorem proving, proof generation, proof verification, and research on self-correcting reasoning systems

Type Reasoning
Context 164K
Reasoning 10/10
Speed 2/10
Status

Available as an open-weight research model; no official DeepSeek-hosted API deployment or public model-specific API pricing verified

View model →
DeepSeek logo
DeepSeek-OCR

DeepSeek-OCR

Local OCR, document digitization, PDF and image parsing, layout-aware markdown conversion, table extraction, figure parsing, and visual-text compression research

Type Multimodal
Context 8K
Reasoning 2/10
Speed 7/10
Multimodal Image input Streaming
Status

Current open-weight model; publicly available for local and self-hosted inference

View model →
DeepSeek logo
DeepSeek-OCR

DeepSeek-OCR 2

Local OCR, scanned-document transcription, layout-aware document parsing, table extraction, and document-to-Markdown workflows

Type Multimodal
Context 8K
Reasoning 2/10
Speed 6/10
Multimodal Image input Streaming
Status

Current open-weight model; self-hosted checkpoint

View model →
DeepSeek logo
DeepSeek-Prover

DeepSeek-Prover-V1

Lean 4 theorem proving, formal mathematics, automated proof generation, and theorem-proving research

Type Reasoning
Reasoning 8/10
Speed 5/10
Status

Open-weight research release; legacy predecessor to DeepSeek-Prover-V1.5 and DeepSeek-Prover-V2

Input No official hosted API pricing; downloadable weights
Output No official hosted API pricing; downloadable weights
View model →

Lean 4 proof completion, formal mathematics research, automated theorem proving, and verifier-guided proof search

Type Reasoning
Context 4K
Reasoning 8/10
Speed 4/10
Fine-tuning
Status

Open-weight, downloadable, legacy/superseded

View model →

Lean 4 theorem proving, formal mathematics, automated proof synthesis, proof-search research, and verifier-guided reasoning

Type Reasoning
Context 164K
Reasoning 9/10
Speed 2/10
Status

Open-weight and downloadable; currently accessible from the official model repository; no first-party hosted API identified for this exact checkpoint

View model →
DeepSeek logo
DeepSeek-Prover-V1.5

DeepSeek-Prover-V1.5-RL

Lean 4 formal theorem proving, mathematical proof generation, proof-code completion, and research on verifier-guided language models

Type Reasoning
Context 4K
Reasoning 8/10
Speed 5/10
Status

Available as a downloadable open-weight model; legacy research release superseded by newer DeepSeek-Prover models

DeepSeek logo
DeepSeek-Prover-V2

DeepSeek-Prover-V2-7B

Lean 4 theorem proving, formal mathematics, proof synthesis, lemma completion, and local proof-search research

Type Reasoning
Context 33K
Reasoning 8/10
Speed 5/10
Status

Available open-weight model

DeepSeek logo
DeepSeek-Prover V1.5

DeepSeek-Prover-V1.5-Base

Lean 4 theorem proving, formal mathematics research, proof completion, proof-search experiments, and open-weight model fine-tuning

Type Reasoning
Context 4K
Reasoning 8/10
Speed 4/10
Status

Legacy open-weight model; downloadable and usable, but superseded by newer DeepSeek-Prover releases

View model →
DeepSeek logo
DeepSeek-R1

DeepSeek-R1

Mathematical reasoning, coding, technical analysis, research, and self-hosted reasoning applications

Type Reasoning
Context 128K
Reasoning 9/10
Speed 4/10
Streaming
Status

Open-weight checkpoint available; original hosted API identity superseded and scheduled for discontinuation

View model →
DeepSeek logo
DeepSeek-R1

DeepSeek-R1-0528-Qwen3-8B

Local mathematical reasoning, coding assistance, research, experimentation, and smaller-scale deployments requiring strong reasoning quality

Type Reasoning
Context 131K
Reasoning 8/10
Speed 7/10
Fine-tuning Streaming
Status

Current open-weight model; downloadable checkpoint

DeepSeek logo
DeepSeek-R1

DeepSeek-R1-Distill-Llama-8B

Local mathematical reasoning, coding assistance, technical question answering, and private self-hosted inference

Type Reasoning
Context 131K
Reasoning 8/10
Speed 7/10
Fine-tuning Streaming
Status

Available as an open-weight model for download and self-hosted or third-party inference

Self-hosted mathematics, coding, research, complex reasoning, and long-form analysis

Type Reasoning
Context 131K
Reasoning 9/10
Speed 4/10
Streaming
Status

Available open-weight model

View model →
DeepSeek logo
DeepSeek-R1

DeepSeek-R1-Distill-Qwen-7B

Local mathematical reasoning, coding assistance, research, and privacy-sensitive self-hosted applications

Type Reasoning
Context 131K
Reasoning 8/10
Speed 7/10
Fine-tuning Streaming
Status

Active open-weight downloadable model

DeepSeek logo
DeepSeek-R1

DeepSeek-R1-Distill-Qwen-14B

Local mathematical reasoning, coding, technical analysis, research, and self-hosted text generation

Type Reasoning
Context 131K
Reasoning 8/10
Speed 6/10
Streaming
Status

Active open-weight model; downloadable and usable through compatible self-hosted inference frameworks

DeepSeek logo
DeepSeek-R1

DeepSeek-R1-Distill-Qwen-32B

Mathematics, programming, complex reasoning, research, and self-hosted inference

Type Reasoning
Context 33K
Reasoning 9/10
Speed 5/10
Fine-tuning Streaming
Status

Available open-weight model

DeepSeek logo
DeepSeek-R1

DeepSeek-R1-Zero

Reasoning research, mathematics, coding experiments, reinforcement-learning studies, open-weight evaluation, and model distillation

Type Reasoning
Context 131K
Reasoning 9/10
Speed 3/10
Fine-tuning Streaming
Status

Open-weight and downloadable; experimental/legacy research model; not a current first-party hosted API model

View model →

Local mathematical reasoning, compact reasoning experiments, educational applications, lightweight coding assistance, and self-hosted inference on limited hardware

Type Reasoning
Context 131K
Reasoning 7/10
Speed 8/10
Fine-tuning Streaming
Status

Active open-weight model; downloadable for self-hosted inference

View model →
DeepSeek logo
DeepSeek-V2

DeepSeek-V2

Local or third-party deployment for general text generation, translation, mathematics, research, and code generation

Type General Purpose
Context 128K
Reasoning 7/10
Speed 7/10
Streaming
Status

Legacy open-weight model; downloadable and usable through local or third-party inference, but not listed in DeepSeek's current first-party hosted API catalog

View model →
DeepSeek logo
DeepSeek-V2

DeepSeek-V2-Lite

Local text generation, Chinese and English language tasks, MoE research, fine-tuning, and efficient self-hosted inference

Type Lightweight
Context 33K
Reasoning 5/10
Speed 7/10
Fine-tuning Streaming
Status

Open-weight and downloadable; legacy self-hosting model

Input No official DeepSeek-hosted API price documented for this exact model; self-hosted weights are available under the DeepSeek Model License.
Output No official DeepSeek-hosted API price documented for this exact model; self-hosted inference costs depend on hardware and serving infrastructure.
View model →
DeepSeek logo
DeepSeek-V2

DeepSeek-V2.5

Open-weight general language generation, coding assistance, code completion, and self-hosted experimentation

Type General Purpose
Context 128K
Reasoning 7/10
Speed 7/10
Tool use Streaming
Status

Legacy open-weight model; the V2.5 series was superseded by newer DeepSeek model families, while the model weights remain available

View model →
DeepSeek logo
DeepSeek-V3

DeepSeek-V3

Open-weight general language generation, coding, long-context text tasks, research, and cost-sensitive third-party inference.

Type General Purpose
Context 128K
Reasoning 8/10
Speed 8/10
Tool use Fine-tuning Streaming
Status

Superseded and no longer current as a first-party hosted API model; open-weight checkpoint remains available for self-hosted and third-party deployment.

Input $0.27 per 1M tokens for cache misses; $0.07 per 1M tokens for cache hits, historical launch pricing
Output $1.10 per 1M tokens, historical launch pricing
View model →
DeepSeek logo
DeepSeek-V3

DeepSeek-V3.1

Open-weight reasoning, coding, tool-calling, long-context analysis, and self-hosted agent systems

Type General Purpose
Context 128K
Reasoning 8/10
Speed 7/10
Tool use Streaming
Status

Legacy hosted API generation; official open-weight release remains available

View model →
DeepSeek logo
DeepSeek-V3.1

DeepSeek-V3.1-Base

Self-hosted research, continued pretraining, fine-tuning, custom inference, and large-scale language or coding workloads

Type General Purpose
Context 131K
Reasoning 8/10
Speed 5/10
Streaming
Status

Available open-weight model; downloadable from Hugging Face

Open-weight deployment, coding assistance, long-context text processing, reasoning workflows, search agents, and terminal-oriented automation

Type General Purpose
Context 128K
Reasoning 8/10
Speed 6/10
Tool use Streaming
Status

Open-weight checkpoint available; dedicated DeepSeek API endpoint retired on 2025-10-15

Input $0.56 per 1M input tokens cache miss; $0.07 per 1M cached input tokens during historical API availability
Output $1.68 per 1M output tokens during historical API availability
View model →
DeepSeek logo
DeepSeek-V3.2

DeepSeek-V3.2-Exp

Long-context text generation, reasoning, coding, research, document analysis, and self-hosted experimentation with sparse attention.

Type General Purpose
Context 164K
Reasoning 8/10
Speed 7/10
Tool use Streaming
Status

Experimental open-weight model; official hosted API identity superseded by DeepSeek-V3.2 on 2025-12-01. Downloadable weights and research code remain available.

Input Historical API pricing: $0.28 per 1M input tokens on cache miss; $0.028 per 1M input tokens on cache hit
Output $0.42 per 1M output tokens

Difficult mathematics, advanced coding, scientific reasoning, long-form analysis, benchmark evaluation, and research deployment

Type Reasoning
Context 164K
Reasoning 10/10
Speed 4/10
Streaming
Status

Retired hosted API; open-weight model remains available for self-hosting and third-party deployment

Input Historical temporary API pricing was the same as DeepSeek-V3.2; no current hosted price because the endpoint expired on 2025-12-15
Output Historical temporary API pricing was the same as DeepSeek-V3.2; no current hosted price because the endpoint expired on 2025-12-15
View model →

Local image understanding, visual question answering, diagram and document analysis, multimodal research, and compact deployments

Type Multimodal
Context 4K
Reasoning 3/10
Speed 7/10
Multimodal Image input
Status

Available open-weight checkpoint; legacy first-generation model

View model →
DeepSeek logo
DeepSeek-VL

DeepSeek-VL-1.3B-Chat

Local image-and-text chat, visual question answering, OCR, document and screenshot understanding, and multimodal research on modest hardware

Type Multimodal
Context 4K
Reasoning 4/10
Speed 7/10
Multimodal Image input Streaming
Status

Publicly available open-weight checkpoint; legacy research model

DeepSeek logo
DeepSeek-VL

DeepSeek-VL-7B-Base

Local visual question answering, image and document understanding, multimodal research, and fine-tuning experiments

Type Multimodal
Context 16K
Reasoning 5/10
Speed 5/10
Multimodal Image input
Status

Open-weight, downloadable, and currently accessible; older model family

DeepSeek logo
DeepSeek-VL

DeepSeek-VL-7B-Chat

Self-hosted image understanding, visual question answering, document and webpage analysis, diagram interpretation, and research prototyping.

Type Multimodal
Context 4K
Reasoning 4/10
Speed 5/10
Multimodal Image input Streaming
Status

Legacy open-weight model; downloadable and self-hostable, with no current first-party hosted API or official inference pricing identified.

DeepSeek logo
DeepSeek-VL2

DeepSeek-VL2

Self-hosted image understanding, OCR, document and chart analysis, visual question answering, and visual grounding research

Type Multimodal
Context 4K
Reasoning 6/10
Speed 4/10
Multimodal Image input
Status

Open-weight and downloadable; currently accessible through the official repository and Hugging Face model page. No first-party hosted API availability verified.

Input No official hosted API input price; downloadable self-hosted model
Output No official hosted API output price; downloadable self-hosted model
DeepSeek logo
DeepSeek-VL2

DeepSeek-VL2-Small

Local or self-hosted visual question answering, OCR, document and chart understanding, image-grounded conversation, and visual grounding

Type Multimodal
Context 4K
Reasoning 5/10
Speed 6/10
Multimodal Image input
Status

Open-weight and currently accessible; released as part of the DeepSeek-VL2 model family

DeepSeek logo
DeepSeek-VL2

DeepSeek-VL2-Tiny

Local visual question answering, OCR, document and chart analysis, image understanding, and visual grounding

Type Multimodal
Context 4K
Reasoning 4/10
Speed 7/10
Multimodal Image input Fine-tuning
Status

Released open-weight model; currently accessible for local deployment

DeepSeek logo
DeepSeekMath

DeepSeekMath-7B-Base

Mathematical reasoning research, local inference, continued pretraining, and task-specific fine-tuning

Type Reasoning
Context 4K
Reasoning 8/10
Speed 6/10
Fine-tuning Streaming
Status

Available open-weight checkpoint; legacy research model

DeepSeek logo
DeepSeekMath

DeepSeekMath-7B-Instruct

Mathematical problem solving, educational assistants, local research, benchmark evaluation, and open-weight reasoning experiments

Type Reasoning
Context 4K
Reasoning 7/10
Speed 6/10
Fine-tuning Streaming
Status

Legacy open-weight model; downloadable and usable for local inference

DeepSeek logo
DeepSeekMath

DeepSeekMath-7B-RL

Open-weight mathematical reasoning, competition-math experiments, local deployment, and research on reinforcement learning for language models

Type Reasoning
Context 4K
Reasoning 8/10
Speed 6/10
Status

Open-weight and downloadable; legacy research model with no official hosted API pricing identified

DeepSeek logo
DeepSeekMoE

DeepSeekMoE 16B Base

Local text completion, open-weight LLM research, domain adaptation, and fine-tuning

Type General Purpose
Context 4K
Reasoning 4/10
Speed 6/10
Fine-tuning
Status

Legacy open-weight model; downloadable and usable for self-hosted deployment

DeepSeek logo
DeepSeek V3

DeepSeek-V3.2

Open-weight reasoning, coding, long-context analysis, tool-using agents, research workflows, and cost-sensitive deployments

Type Reasoning
Context 131K
Reasoning 8/10
Speed 7/10
Tool use Streaming
Status

Legacy open-weight model; former DeepSeek API aliases deepseek-chat and deepseek-reasoner were scheduled for discontinuation on 2026-07-24

Input $0.028 per 1M tokens cached; $0.28 per 1M tokens cache miss during official API availability
Output $0.42 per 1M tokens during official API availability
View model →
DeepSeek logo
DeepSeek V3.2

DeepSeek-V3.2-Exp-Base

Long-context architecture research, self-hosted inference, continued pretraining, and custom model adaptation

Type General Purpose
Context 164K
Reasoning 8/10
Speed 6/10
Streaming
Status

Experimental, downloadable open-weight checkpoint; superseded by DeepSeek-V3.2

Long-context reasoning, coding, agent workflows, and cost-sensitive API applications

Type Lightweight
Context 1M
Reasoning 8/10
Speed 9/10
Tool use Streaming
Status

Retired; legacy API identifier temporarily routed to DeepSeek-V4.1-Flash from September 10, 2026

Input $0.14 per 1 million input tokens
Output $0.28 per 1 million output tokens
View model →

Self-hosted language-model research, custom post-training, domain adaptation, and large-context text generation

Type General Purpose
Context 1.05M
Reasoning 7/10
Speed 6/10
Status

Current downloadable open-weight base checkpoint; no Hugging Face Inference Provider deployment listed

View model →

Image understanding, screenshot and chart analysis, multimodal coding agents, visual tool-use workflows, and text-plus-image reasoning

Type Multimodal
Reasoning 8/10
Speed 8/10
Multimodal Image input Tool use
Status

Retired as an independent model on 2026-09-10; legacy API identifier temporarily routes requests to DeepSeek-V4.1-Flash

Input $0.15 per 1M cache-miss input tokens off-peak or $0.30 peak; $0.003 off-peak or $0.006 peak for cache-hit input when using the current routed Flash pricing
Output $0.60 per 1M output tokens off-peak or $1.20 peak when using the current routed Flash pricing
View model →
DeepSeek logo
DeepSeek V4

DeepSeek-V4-Pro

Complex reasoning, coding agents, long-context analysis, tool-using workflows, and large document or codebase processing

Type General Purpose
Context 1M
Reasoning 9/10
Speed 6/10
Tool use Streaming
Status

Deprecated for independent serving; deepseek-v4-pro API requests are routed to DeepSeek-V4.1-Flash until V4.1-Pro launches

Input Official V4-Pro list price: US$0.66 per 1M cache-miss input tokens off-peak and US$1.32 peak; US$0.022 per 1M cache-hit input tokens off-peak and US$0.044 peak. DeepSeek announced that routed requests use V4.1-Flash rates.
Output Official V4-Pro list price: US$1.98 per 1M output tokens off-peak and US$3.96 peak. DeepSeek announced that routed requests use V4.1-Flash rates.
View model →

Research, continued pretraining, fine-tuning, foundation-model evaluation, and custom large-scale inference

Type General Purpose
Context 1.05M
Reasoning 8/10
Speed 4/10
Status

Current open-weight base checkpoint; downloadable under the MIT License

View model →
DeepSeek logo
DeepSeek V4.1

DeepSeek-V4.1-Flash

Low-cost, high-throughput reasoning and coding, long-context analysis, agentic workflows, tool calling, and text-plus-image understanding.

Type Lightweight
Context 1.05M
Reasoning 9/10
Speed 10/10
Multimodal Image input Tool use
Status

Current and available through the DeepSeek API; the canonical API identifier is deepseek-flash.

Input $0.15 per 1M input tokens off-peak or $0.30 peak for cache misses; $0.003 off-peak or $0.006 peak for cache hits.
Output $0.60 per 1M output tokens off-peak or $1.20 peak.
View model →

Local research, image understanding, visual question answering, multimodal prototyping, and lightweight text-to-image experimentation

Type Multimodal
Context 4K
Reasoning 4/10
Speed 7/10
Multimodal Image input Media output
Status

Available as an open-weight research model; superseded in the Janus series by newer Janus-Pro variants but still downloadable and usable.

Input No official hosted API pricing; self-hosted model weights
Output No official hosted API pricing; self-hosted model weights
View model →
DeepSeek logo
Janus-Pro

Janus-Pro-1B

Local multimodal research, image understanding, visual question answering, and compact text-to-image experimentation

Type Multimodal
Context 4K
Reasoning 5/10
Speed 7/10
Multimodal Image input Media output
Status

Available as an open-weight downloadable model; no official hosted inference provider is listed for the exact checkpoint.

DeepSeek logo
Janus-Pro

Janus-Pro-7B

Local image understanding, text-to-image generation, and unified multimodal research

Type Multimodal
Context 4K
Reasoning 5/10
Speed 4/10
Multimodal Image input Media output
Status

Current open-weight model; downloadable and usable for local deployment

Local research, visual question answering, image interpretation, and compact text-to-image experimentation

Type Multimodal
Context 4K
Reasoning 5/10
Speed 7/10
Multimodal Image input Media output
Status

Available as an open-weight downloadable checkpoint; no official hosted inference API identified

Input No official hosted API pricing
Output No official hosted API pricing
View model →