DeepSeek V3 family
General-purpose MoE base and chat releases that underpin the separate R1 reasoning family; the original V3 card documents 671B total and 37B active parameters.
Score calculation
(2 + 3 + 8 + 1 + 7) ÷ 5 = 4.2
5 of six dimensions rated. Unrated dimensions are excluded from the average.
Full scoring method ↗Setup details
DeepSeek V3 and V3.2 open checkpoints, the released base lineage underlying R1; distinct from the newer DeepSeek V4 profile.
What fits your needs
DeepSeek V3 and V3.2 open checkpoints, the released base lineage underlying R1; distinct from the newer DeepSeek V4 profile.
Good fit
General-purpose MoE base and chat releases that underpin the separate R1 reasoning family; the original V3 card documents 671B total and 37B active parameters.
Tradeoffs to expect
Tradeoffs
MIT for the official V3 repository and weights, as stated in the model card; check V3.2 separately. MIT weights, source code, and detailed architecture/training reports make V3 substantially inspectable and reproducible. Its data sources and permissions remain unresolved; code/weight licensing is not training data permission.
How to switch
Local setup
The 671B total-parameter BF16 checkpoint is server-scale. Official serving instructions target distributed GPU systems; active parameters alone do not define memory needs.
Practical questions
Weights and license
Fetch the exact checkpoint revision and read its own license; sibling and distilled variants may inherit different terms.
Hardware
The 671B total-parameter BF16 checkpoint is server-scale. Official serving instructions target distributed GPU systems; active parameters alone do not define memory needs.
Training data
Treat the weight license and public web/data availability separately from evidence of permission for the training data.
This is a practices and disclosure score for the named release, not a request-level energy or emissions estimate.
See the evidence
The V3 report gives a training-compute estimate in H800 GPU-hours for its final training run; this is not a direct electricity, emissions, or water measurement. The rubric rewards documented reporting and controls; it does not infer high consumption from missing measurements.
1 + 1 reporting + 0 electricity + 0 controls = 2/10. This grades documented practices and transparency.
No comparable per-request electricity, emissions, water, regional serving mix, or independently reviewed model footprint is established in the cited sources.
- Reporting1/3
- The V3 report gives a training-compute estimate in H800 GPU-hours for its final training run; this is not a direct electricity, emissions, or water measurement.
- Electricity sourcing0/3
- No model-specific clean-electricity sourcing or hourly power evidence is published in the reviewed sources; zero means unestablished, not dirty electricity.
- Resource controls0/3
- No model-specific documented control or measured physical resource reduction is established by the reviewed release materials. Public weights or smaller variants alone do not prove reduced resource use.
Water disclosure
Water reporting has not been assessed for this choice.
The release provides some training-process information, but not a complete permission ledger for inherited and later training data.
See the evidence
The technical report describes a broad data pipeline and training stages, but does not disclose a full source-level training data and permission ledger for the inherited pretraining mixture. A public weight license governs use of those weights; it is not proof that every training source was permitted.
1 base + 2 source transparency + 0 documented permission + 0 creator controls = 3/10. Weight licensing is not treated as training data permission.
Complete source lists, source-specific rights, inherited-base provenance, synthetic-data generation permissions, and creator removal or retraining remedies remain unestablished.
- Source transparency2/3
- The technical report describes a broad data pipeline and training stages, but does not disclose a full source-level training data and permission ledger for the inherited pretraining mixture.
- Documented permission0/4
- A model-weight license governs the released checkpoint, not automatically the training data. The reviewed release doesn’t confirm affirmative rights for all inherited and post-training material.
- Creator controls0/2
- The reviewed model documentation doesn’t show creator permission covering all training data, opt-out, removal, or retraining remedy for this model family.
Privacy depends on where these released weights are run; the model family has no single prompt-retention or training route.
See the evidence
The official release makes weights available for separate deployments, but it does not define one inference host or retention policy for every user. Local and hosted operators may handle prompts differently, so a single numerical privacy grade would misstate this model-level entry.
The user-selected host, region, telemetry, logging, extensions, and provider training terms are outside the checkpoint release and require a route-specific review.
MIT weights, source code, and detailed architecture/training reports make V3 substantially inspectable and reproducible. Its data sources and permissions remain unresolved; code/weight licensing is not training data permission.
See the evidence
MIT weights, source code, and detailed architecture/training reports make V3 substantially inspectable and reproducible. Its data sources and permissions remain unresolved; code/weight licensing is not training data permission. This openness rating concerns artifacts and reproducibility only; it does not assess whether the original training data was authorized.
MIT for the official V3 repository and weights, as stated in the model card; check V3.2 separately. Model weights and documentation are available, but limitations described above reduce source-level reproducibility.
The checked release and license apply to the named checkpoints; sibling variants, future versions, and the underlying training data may have different terms or disclosures.
DeepSeek’s reviewed model, privacy and user-term materials don’t confirm employment conditions, a worker grievance route or labor assurance for its AI supply chain.
See the evidence
The sources reviewed provide product, terms, or corporate information but no qualifying disclosure about conditions for data annotators, evaluators, moderators, or infrastructure workers. This is a disclosure gap within the checked materials, not evidence that workers are mistreated or that no labor policy exists.
1 base + 0 conditions + 0 worker voice/remedy + 0 evidence/follow-through = 1/10. This grades documented safeguards, not proven mistreatment.
Worker counts, locations, pay, labor protections, worker voice mechanisms, supplier coverage, and independent audit or remediation outcomes were not established in the sources reviewed.
- Pay & conditions0/3
- Reviewed sources provide no qualifying disclosure of worker conditions; this zero is limited to the checked material and is not an exhaustive claim.
- Worker voice & remedy0/3
- No qualifying worker-voice mechanism or documented worker grievance outcomes were disclosed in the reviewed material.
- Evidence & follow-through0/3
- No independent labor audit results, worker-level corroboration, or remediation outcomes were disclosed in the reviewed material.
No DeepSeek military customer, contract or deployment is identified in the reviewed official sources.
See the evidence
DeepSeek’s reviewed official terms, privacy material and model disclosures do not identify a defense customer, military contract or deployment. Public model artifacts can be reused downstream, but that doesn’t confirm DeepSeek’s own military relationship or any customer-specific safeguard.
7 base + 0 safeguard adjustment = 7/10 (0 of 9 safeguard points; every 3 adds 1). No military relationship was established in the cited sources. This ranks above known defense ties without claiming a verified clean record.
Complete contract texts, undisclosed customers, end-use restrictions, actual workload deployment, and independent compliance or outcome reports are unavailable in the reviewed public material.
- Use restrictions0/4
- The reviewed terms or policy material contains no qualifying military-use restriction beyond general lawfulness/export compliance; this zero is limited to the reviewed source set.
- Enforcement & oversight0/3
- No military-specific review, enforcement, or remedy mechanism was disclosed in the sources reviewed.
- Relationship disclosure0/2
- No military-specific relationship or military-specific transparency policy was disclosed in the materials reviewed; this is a scoped disclosure finding.
DeepSeek’s model disclosures and user terms reviewed here do not identify a military contract, named defense customer, or military deployment. Its open model artifacts permit downstream uses, but that doesn’t show DeepSeek’s own military relationship.
Military rubric ↗DeepSeek is not named in the two official BNC campaign lists checked. This is a limited list check, not BDS approval.
See the evidence
The official BNC consumer boycott/pressure guide and No Tech campaign page do not name DeepSeek among their current targets. BDS uses selective campaigns, so absence from these lists is neutral and should not be read as a clean record or approval.
This check covers only named targets in the two cited BNC campaigns; it does not cover all BDS campaigns, local groups, investments, suppliers, or downstream hosts. Lists and ownership may change.