Overview

Identity

Identity block not returned by the API for this entity.

Openness

Openness not classified yet — no sourced evidence to place this model in the ontology.

Key facts

Status: Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted
Version: Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted
Official page: chat.deepseek.com
Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted
Paper: huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf
Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted
Repository: huggingface.co/collections/deepseek-ai/deepseek-v4
Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted

Architecture

Architecture: Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted
Parameters: Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted
Active parameters: Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted
Mixture of experts: Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted

Capabilities

Modalities

Modalities: text
Input: text
Output: text

Capabilities

Tool calling
Yes
DeepSeek — site & API docs · T2
Structured output
Unavailable
Reasoning
Yes
DeepSeek — site & API docs · T2
Vision
Unavailable
Audio
Unavailable
Fine-tuning available
Unavailable

Context window: Source:DeepSeek — site & API docsT2observed 10 h agomediumLLM-extracted

Hardware fit37

Estimated

1 of 37 device × quantization combinations fit.

Run locally: your machine →

Estimated hardware fit
Hardware	Quantization	Device memory	Est. memory	Fits
NVIDIA DGX B200	4bit	1,440 GB	920.5 GB est.	Yes
Apple M3 Ultra	4bit	—	920.5 GB est.	No
Mac Studio (Apple M5 Ultra)	4bit	—	920.5 GB est.	No
Apple M2 Ultra	4bit	—	920.5 GB est.	No
Apple M1 Ultra	4bit	—	920.5 GB est.	No
Apple M3 Max	4bit	—	920.5 GB est.	No
Apple M4 Max	4bit	—	920.5 GB est.	No
Mac Studio (Apple M5 Max)	4bit	—	920.5 GB est.	No
MacBook Pro (Apple M5 Max)	4bit	—	920.5 GB est.	No
Apple M2 Max	4bit	—	920.5 GB est.	No
Apple M1 Max	4bit	—	920.5 GB est.	No
Apple M4 Pro	4bit	—	920.5 GB est.	No
Mac mini (Apple M5 Pro)	4bit	—	920.5 GB est.	No
MacBook Pro (Apple M5 Pro)	4bit	—	920.5 GB est.	No
Apple M3 Pro	4bit	—	920.5 GB est.	No
Apple M1 Pro	4bit	—	920.5 GB est.	No
Apple M2 Pro	4bit	—	920.5 GB est.	No
Apple M4	4bit	—	920.5 GB est.	No
iMac (Apple M4)	4bit	—	920.5 GB est.	No
Mac mini (Apple M6)	4bit	—	920.5 GB est.	No
MacBook Air (Apple M5)	4bit	—	920.5 GB est.	No
MacBook Pro (Apple M5)	4bit	—	920.5 GB est.	No
Apple M2	4bit	—	920.5 GB est.	No
Apple M3	4bit	—	920.5 GB est.	No
Apple M1	4bit	—	920.5 GB est.	No
NVIDIA GeForce RTX 3090	4bit	24 GB	920.5 GB est.	No
NVIDIA GeForce RTX 4090	4bit	24 GB	920.5 GB est.	No
NVIDIA GeForce RTX 5090	4bit	32 GB	920.5 GB est.	No
NVIDIA A100 80GB	4bit	80 GB	920.5 GB est.	No
NVIDIA H100 SXM	4bit	80 GB	920.5 GB est.	No
NVIDIA H100 NVL	4bit	94 GB	920.5 GB est.	No
NVIDIA DGX Spark	4bit	128 GB	920.5 GB est.	No
NVIDIA H200	4bit	141 GB	920.5 GB est.	No
NVIDIA H200 NVL	4bit	141 GB	920.5 GB est.	No
NVIDIA B200	4bit	180 GB	920.5 GB est.	No
AMD Instinct MI300X	4bit	192 GB	920.5 GB est.	No
AMD Instinct MI325X	4bit	256 GB	920.5 GB est.	No

Assumptions (6)

Estimated, not measured: weights = parameters × bytes/param × 1.15 runtime overhead.
bytes/param: 4bit = 0.5, 8bit = 1.0, fp16 = 2.0 (uniform quantization, no per-layer exceptions).
KV cache approximated at 0.5 GB per 8 192 tokens of context, independent of architecture (GQA/MLA models need less).
A model 'fits' when the estimate is at most the device memory minus 2 GB reserved for the OS and framework.
Mixture-of-experts models are estimated on total parameters (all experts must be resident); active parameters are ignored.
Device memory uses the largest configuration when several are listed (e.g. Apple silicon tiers).

Timeline1

Full timeline →

11 Sept 2026
New modelModelDeepSeek-V4DeepSeek
New model: DeepSeek-V4 (DeepSeek)
deepseek
api-docs.deepseek.com

Change history18

Temporal, append-only claims: a new observation closes the previous claim instead of overwriting it. Rewind the record with the as-of picker.

18 claims · 18 properties

Familyfamily1

Claim history for Family
Value	Valid from → to	Status	Source	Confidence	Extractor
DeepSeek-V4	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Versionversion1

Claim history for Version
Value	Valid from → to	Status	Source	Confidence	Extractor
Preview	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Statusstatus1

Claim history for Status
Value	Valid from → to	Status	Source	Confidence	Extractor
preview	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Opennessopenness1

Claim history for Openness
Value	Valid from → to	Status	Source	Confidence	Extractor
open-weights	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Architecturearchitecture1

Claim history for Architecture
Value	Valid from → to	Status	Source	Confidence	Extractor
transformer decoder with Token-wise compression and DSA (DeepSeek Sparse Attention)	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Parametersparameter_count1

Claim history for Parameters
Value	Valid from → to	Status	Source	Confidence	Extractor
1.6T	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Active parametersactive_parameter_count1

Claim history for Active parameters
Value	Valid from → to	Status	Source	Confidence	Extractor
49B	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Mixture of expertsis_moe1

Claim history for Mixture of experts
Value	Valid from → to	Status	Source	Confidence	Extractor
Yes	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Context windowcontext_length1

Claim history for Context window
Value	Valid from → to	Status	Source	Confidence	Extractor
1M tokens	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Modalitiesmodalities1

Claim history for Modalities
Value	Valid from → to	Status	Source	Confidence	Extractor
text	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Input modalitiesmodalities_input1

Claim history for Input modalities
Value	Valid from → to	Status	Source	Confidence	Extractor
text	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Output modalitiesmodalities_output1

Claim history for Output modalities
Value	Valid from → to	Status	Source	Confidence	Extractor
text	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Official pageofficial_url1

Claim history for Official page
Value	Valid from → to	Status	Source	Confidence	Extractor
https://chat.deepseek.com	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Paperpaper_url1

Claim history for Paper
Value	Valid from → to	Status	Source	Confidence	Extractor
https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Repositoryrepository_url1

Claim history for Repository
Value	Valid from → to	Status	Source	Confidence	Extractor
https://huggingface.co/collections/deepseek-ai/deepseek-v4	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Reasoningreasoning1

Claim history for Reasoning
Value	Valid from → to	Status	Source	Confidence	Extractor
Yes	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Safety notessafety_notes1

Claim history for Safety notes
Value	Valid from → to	Status	Source	Confidence	Extractor
deepseek-chat & deepseek-reasoner will be fully retired and inaccessible after Jul 24th, 2026, 15:59 (UTC Time).	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Tool callingtool_calling1

Claim history for Tool calling
Value	Valid from → to	Status	Source	Confidence	Extractor
Yes	11 Sept 2026 20:52 UTC → current	current	DeepSeek — site & API docsT2	medium	llm

Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →

Provenance

Attributed facts

18

Source tiers

T218

Freshest observation

10 h ago

Conflicts

None

Source documents 1

Source documents
Source	Document	Type	Tier	Last observed	Snapshots
DeepSeek — site & API docs	api-docs.deepseek.com/news/news260424	news	T1· Official	10 h ago	1

Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.

Data quality (55/100) measures how well AI Atlas knows this entity — completeness, primary-source ratio, freshness, conflicts — never how good the model is. Methodology →