DEVELOPER REFERENCE · VERIFIED MODEL EVIDENCE

DeepSeek
Compare 21 source-verified model records, repository licenses, published weight formats and dated evidence. No overall model ranking.
Evidence statusSource-first verified
Latest record check2026-11-13
Latest record full verification2026-11-10
DIRECT ANSWER · SOURCE-OBSERVED REGISTRY
What DeepSeek model evidence is available?
Open Model Weights currently tracks 21 field-verified DeepSeek model records. In this indexed sample, the recorded modalities are 16 Text; 5 Image + text → text. The published repository license labels include 17 mit; 2 other; 1 apache-2.0; 1 Unknown / not declared. The listed weight formats include Safetensors (21).
Coverage describes the model repositories indexed here, not every model released by the publisher. A checked source, an unknown field and an unchanged source revision are different states. No missing value is treated as a negative answer.
21verified model records
2recorded modalities
4distinct license labels
21model repository links
DEVELOPER HQ / PRIMARY ORGANIZATION LOCATION
🇨🇳
Primary organization locationHangzhou, China
Hangzhou DeepSeek Artificial Intelligence Co., Ltd.
This location describes the developer or primary organization. It is not the nationality of a model, dataset, contributor, training run or infrastructure.
DeepSeek publishes downloadable language, reasoning and document-understanding checkpoints, alongside hosted model services. Official DeepSeek repositories describe distinct R1 and V3 model families, distillation variants and their license documents.
Registry coverage is repository-specific. An R1 reasoning model, a V3 chat/base checkpoint, a Qwen-derived distilled checkpoint and DeepSeek-OCR do not have the same task, deployment footprint or upstream license ancestry.
01R1 reasoning versus the V3 family
DeepSeek-R1 is a reasoning-oriented model family; V3 records provide separate language-model checkpoints and revisions. Architecture, intended prompts and serving constraints may differ. A model name or MoE active-parameter count cannot substitute for verifying the actual artifact and supported inference stack.
02Why distillation ancestry matters
DeepSeek's R1 repository explains that R1-Distill-Qwen variants derive from Qwen models, while other distillations have different base-model provenance. The derivative's license must be interpreted alongside any applicable upstream terms; identical downstream branding does not erase source lineage.
03Vision and document tasks
The DeepSeek-OCR family focuses on optical document understanding rather than general text-only completion. DeepSeek-OCR-2 is separately indexed as a multimodal model with an Apache-2.0 license label; compare it against text-only R1/V3 models only within explicitly defined document workflows.
04Care with code versus model licensing
The official DeepSeek-V3 README explicitly distinguishes the MIT-licensed code repository from the separately licensed model weights. R1 documents MIT weight licensing for its released checkpoints. These facts do not establish that every DeepSeek checkpoint and historical derivative shares the same terms; check the recorded license file.
Publisher background is attributed to official sources. Model counts and license classifications come from OMW repository evidence; no model benchmark or processing location is inferred.
Textmit
Reasoning-oriented language model
Start with the DeepSeek-R1 record for reasoning-oriented tasks and inspect the available runtime and weight format.
Textmit
V3-family language model
Open DeepSeek-V3.2 for its exact V3-family revision and its current source-declared terms.
Image + text → textapache-2.0
Document and image understanding
Inspect DeepSeek-OCR-2 for multimodal text extraction rather than treating OCR as a chat benchmark.
Textmit
Qwen-based distilled reasoning
Review R1-Distill-Qwen-32B with attention to both derivation and release-specific licensing.
Textother
Code-oriented checkpoint
Check the separately listed Coder-V2-Lite-Instruct model and its custom or unresolved license before deployment.
Compare models →Explore weight-memory estimates →Understand licensing →
2026-10-07 · RuntimeThe October 5 serving release adds a GPU-resident weight-cache daemon, experimental initialized-engine snapshots, new speculative-decoding paths and a broad set of DeepSeek V4.1 Flash optimizations.
FIELD-BY-FIELD COMPARISON · PRIMARY LINKS
Model evidence at a glance.
Compare the exact entries in our registry. Parameter values are taken from the source-checked record; a license label is not a legal opinion, and recognized weight files are not an inference-speed benchmark. Open each record for the underlying citations and observed history.
Compare two to four models →Explore weight memory →Browse observed history →
LICENSE & COVERAGE · NOT LEGAL ADVICE
How do the recorded licenses differ?
Different models from the same developer can use different license terms. The distribution below describes the labels and field-level classifications observed in the listed repositories; it does not automatically establish permissions for a specific deployment.
License labels
- mit17 of 21
- other2 of 21
- apache-2.01 of 21
- Unknown / not declared1 of 21
Commercial-use classifications
- allowed18 of 21
- unknown3 of 21
Model modalities
- Text16 of 21
- Image + text → text5 of 21
EVIDENCE METHOD · TRANSPARENT LIMITS
How this developer profile is assembled.
We aggregate the current source-checked model records, rather than publish an unverified corporate biography. Repository metadata, model cards, license files and recognized weight artifacts supply the facts where available. The record-check date does not assert that a model or its license changed on that date.
Source-derived runtime signals are not hardware benchmarks. Weight-only memory estimates omit runtime overhead, activations and other deployment requirements. Model-specific uncertainties remain visible in the individual evidence records.
Evidence policy →Full methodology →Independent measurements →
QUESTIONS · DIRECT ANSWERS
Frequently asked questions about DeepSeek models.
Which DeepSeek models are covered?
Open Model Weights currently lists 21 field-verified records for DeepSeek. Their recorded model modalities are: 16 Text; 5 Image + text → text. This is a registry sample, not an exhaustive publisher catalog.
Which licenses are recorded for DeepSeek models?
In the 21 indexed records, repository license labels are: 17 mit; 2 other; 1 apache-2.0; 1 Unknown / not declared. License labels do not replace the model-specific terms or establish rights for every use case.
Where can I verify DeepSeek model information?
Each indexed model links to its official source repository and a field-level Open Model Weights profile. The most recent record checks shown here reach 2026-11-13; checking a repository does not imply that its contents changed.
Is DeepSeek-R1 available as downloadable weights?
Yes. DeepSeek provides an official R1 repository and downloadable Hugging Face checkpoints. Reproducing a hosted service's behavior remains a separate engineering task.
Are all DeepSeek models licensed under MIT?
No. DeepSeek-V3 documentation separates code and model licenses, and the registry includes custom or unresolved terms for some checkpoints. Verify rights at the exact model revision.
Is an R1-Distill-Qwen checkpoint equivalent to DeepSeek-R1?
No. A distilled model uses a different base checkpoint and parameter scale. DeepSeek's own R1 documentation identifies upstream Qwen ancestry for several variants.
COMPLETE INDEXED SELECTION
All 21 indexed model profiles.
163,840 tokens
mit
Field-verified
DeepSeek-V4-Flash-0731
Text
1,048,576 tokens
mit
Field-verified
DeepSeek-OCR
Image + text → text
8,192 tokens
mit
Field-verified
DeepSeek-V4.1-Flash
Image + text → text
1,048,576 tokens
mit
Field-verified
163,840 tokens
mit
Field-verified
163,840 tokens
Unknown / not declared
Field-verified
1,048,576 tokens
mit
Field-verified
DeepSeek-R1-0528-Qwen3-8B
Text
131,072 tokens
mit
Field-verified
DeepSeek-R1-Distill-Qwen-1.5B
Text
131,072 tokens
mit
Field-verified
DeepSeek V4.1 Flash 1017
Image + text → text
1,048,576 tokens
mit
Field-verified
DeepSeek-R1-Distill-Qwen-32B
Text
131,072 tokens
mit
Field-verified
1,048,576 tokens
mit
Field-verified
DeepSeek-OCR-2
Image + text → text
8,192 tokens
apache-2.0
Field-verified
163,840 tokens
mit
Field-verified
163,840 tokens
mit
Field-verified
DeepSeek-R1-Distill-Qwen-14B
Text
131,072 tokens
mit
Field-verified
DeepSeek-V4-Flash-DSpark
Text
1,048,576 tokens
mit
Field-verified
DeepSeek-V4-Flash-Vision-Exp
Image + text → text
1,048,576 tokens
mit
Field-verified
deepseek-coder-7b-instruct-v1.5
Text
4,096 tokens
other
Field-verified
163,840 tokens
mit
Field-verified
DeepSeek-Coder-V2-Lite-Instruct
Text
163,840 tokens
other
Field-verified