DeepSeek
DeepSeek has 10 models in Epoch AI's notable models list, 7 of them released since January 2025; the largest training run with a compute estimate is DeepSeek-V4-Pro, at 9.7e24 FLOP. It has published no frontier safety framework, and none of its 4 indexed releases since January 2025 had a safety card on the day of release.
- 9.7e24 FLOP largest training run, DeepSeek-V4-Pro, 2026 Notable AI Models Epoch AI Inclusion is editorial ("notability criteria"), not exhaustive. Compute figures are frequently estimates; check the per-row confidence and notes columns before treating any value as measured. Cite Epoch AI, "Notable AI Models". https://epoch.ai/data/ai-models. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/developers/deepseek/.
- 7 notable models released since January 2025
- 0 of 4 releases since January 2025 with a system card on the day Release Safety Disclosure Index This site Our own reading of the system card, model card or safety report each developer published for each release Epoch AI lists from 2025, joined to Epoch's list. Every cell is what a developer says about its own model under its own framework: whether a card came out with the release, whether it reports dangerous-capability results, which government institutes it names as pre-release testers, and whether it says a risk threshold was reached. None of it is verified. A release Epoch adds after the last review is shown as awaiting review. Cite AI Safety Tracker, "Release Safety Disclosure Index". https://aisafetytracker.org/alignment/#releases. Retrieved 18 October 2026. Notable AI Models Epoch AI Inclusion is editorial ("notability criteria"), not exhaustive. Compute figures are frequently estimates; check the per-row confidence and notes columns before treating any value as measured. Cite Epoch AI, "Notable AI Models". https://epoch.ai/data/ai-models. Retrieved 18 October 2026 via AI Safety Tracker, https://aisafetytracker.org/developers/deepseek/.
- 0 sites it owns in Epoch's data-centre hub
Models
Every model Epoch AI judges notable from DeepSeek, by year, with the largest training run each year where Epoch has a compute estimate.
| Year | Models | Models, relative | Largest run that year |
|---|---|---|---|
| 2024 | 3 | DeepSeek-V3, 3.3e24 FLOP | |
| 2025 | 4 | DeepSeek-R1 (May 2025), 4.0e24 FLOP | |
| 2026 | 3 | DeepSeek-V4-Pro, 9.7e24 FLOP |
Most recent
| Published | Model | Domain | Released as | Training compute |
|---|---|---|---|---|
| 2026-09-09 | DeepSeek V4.1 Flash | Language | Open weights | 4.3e24 |
| 2026-04-24 | DeepSeek-V4-Pro | Language | Open weights | 9.7e24 |
| 2026-04-24 | DeepSeek-V4-Flash | Language | Open weights | 2.5e24 |
| 2025-11-27 | DeepSeekMath-V2 | Language | Open weights | - |
| 2025-05-28 | DeepSeek-R1 (May 2025) | Language | Open weights | 4.0e24 |
| 2025-03-24 | DeepSeek-V3 (Mar 2025) | Language | Open weights | 3.3e24 |
| 2025-01-20 | DeepSeek-R1 | Language | Open weights | 3.5e24 |
| 2024-12-24 | DeepSeek-V3 | Language | Open weights | 3.3e24 |
| 2024-09-06 | DeepSeek-V2.5 | Language | Open weights | 1.8e24 |
| 2024-06-17 | DeepSeek-Coder-V2 236B | Language | Open weights | 1.3e24 |
Notability is Epoch's editorial judgement, and compute figures are its estimates, most of them not disclosed by the developer. Notable AI Models Epoch AI Inclusion is editorial ("notability criteria"), not exhaustive. Compute figures are frequently estimates; check the per-row confidence and notes columns before treating any value as measured. Cite Epoch AI, "Notable AI Models". https://epoch.ai/data/ai-models. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/developers/deepseek/. Notable AI Models Epoch AI Inclusion is editorial ("notability criteria"), not exhaustive. Compute figures are frequently estimates; check the per-row confidence and notes columns before treating any value as measured. Cite Epoch AI, "Notable AI Models". https://epoch.ai/data/ai-models. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/developers/deepseek/.
Safety framework and release disclosures
The frontier safety framework DeepSeek has published, and what it published at each release since January 2025.
No published frontier safety framework.
Listed as an absence. DeepSeek releases open-weight models at or near the frontier and has published no safety framework, and is not among the Seoul signatories. The linked document is the commitment list it is absent from.
- 0 of 4 with a system card on the day
- 4 that no card names
- 0 naming a government institute that tested before release
- 0 stated to have reached a threshold in its own framework
| Released | Model | Card | Results | Government tested | Threshold |
|---|---|---|---|---|---|
| 2026-09-09 | DeepSeek V4.1 Flash Its technical report gives offensive-cyber benchmark scores as capabilities, with no risk assessment. | none | - | - | - |
| 2026-04-24 | DeepSeek-V4-Pro Released with a technical report on capability. DeepSeek publishes no model card or safety report. | none | - | - | - |
| 2026-04-24 | DeepSeek-V4-Flash Released with a technical report on capability. DeepSeek publishes no model card or safety report. | none | - | - | - |
| 2025-01-20 | DeepSeek-R1 Released with a technical report on capability. DeepSeek publishes no model card or safety report. | none | - | - | - |
Every entry is DeepSeek's own statement about its own models under its own framework. The rules behind each column are on the alignment page. Frontier Safety Framework Index This site Our own reading of each developer's published safety framework and model cards. Every field is what the developer says about itself: these are self-assessments under self-written frameworks with no common scale and no external audit, and they are indexed here for exactly that reason. Not a ranking, and not a verification. Cite AI Safety Tracker, "Frontier Safety Framework Index". https://aisafetytracker.org/alignment/#frameworks. Retrieved 18 October 2026. Release Safety Disclosure Index This site Our own reading of the system card, model card or safety report each developer published for each release Epoch AI lists from 2025, joined to Epoch's list. Every cell is what a developer says about its own model under its own framework: whether a card came out with the release, whether it reports dangerous-capability results, which government institutes it names as pre-release testers, and whether it says a risk threshold was reached. None of it is verified. A release Epoch adds after the last review is shown as awaiting review. Cite AI Safety Tracker, "Release Safety Disclosure Index". https://aisafetytracker.org/alignment/#releases. Retrieved 18 October 2026. Notable AI Models Epoch AI Inclusion is editorial ("notability criteria"), not exhaustive. Compute figures are frequently estimates; check the per-row confidence and notes columns before treating any value as measured. Cite Epoch AI, "Notable AI Models". https://epoch.ai/data/ai-models. Retrieved 18 October 2026 via AI Safety Tracker, https://aisafetytracker.org/developers/deepseek/.
Its models against the compute thresholds
How many of DeepSeek's models with a compute estimate sit at or above each training-compute threshold written into an instrument, and the first to cross it.
| Instrument | Threshold | Status | Its models above | First of them to cross |
|---|---|---|---|---|
| Regulation (EU) 2024/1689, Article 51(2) European Union | 1e25 | in force | 0 | - |
| SB 53, Transparency in Frontier Artificial Intelligence Act United States (California) | 1e26 | in force | 0 | - |
| Enforcement Decree of the AI Basic Act, Article 23 South Korea | 1e26 | in force | 0 | - |
| Executive Order 14110, section 4.2 United States (federal) | 1e26 | revoked | 0 | - |
| SB 1047, Safe and Secure Innovation for Frontier AI Models Act United States (California) | 1e26 | vetoed | 0 | - |
| RAISE Act, Responsible AI Safety and Education Act United States (New York) | 1e26 | enacted | 0 | - |
Arithmetic on Epoch's compute estimates against the number in each instrument, not a legal finding about any model. The full timeline is on the capability page. Compute Threshold Index This site Training-compute thresholds written into law, executive action or published policy, with the text that sets each one. Whether a given model crosses a threshold here is arithmetic on Epoch's compute estimates, most of which are estimates rather than disclosures, and is not a legal finding about any model or developer. Cite AI Safety Tracker, "Compute Threshold Index". https://aisafetytracker.org/capability/#thresholds. Retrieved 23 October 2026. Notable AI Models Epoch AI Inclusion is editorial ("notability criteria"), not exhaustive. Compute figures are frequently estimates; check the per-row confidence and notes columns before treating any value as measured. Cite Epoch AI, "Notable AI Models". https://epoch.ai/data/ai-models. Retrieved 23 October 2026 via AI Safety Tracker, https://aisafetytracker.org/developers/deepseek/.
Datacentres
Sites in Epoch AI's data-centre hub that DeepSeek owns or is named as using, largest planned IT power first.
Epoch's hub names DeepSeek at no site, as owner or as user.
Power is Epoch's estimate from imagery and permits, IT power where it states one. Confidence is Epoch's own label for each owner or user. AI Data Centers Epoch AI The largest AI sites Epoch has found, mostly in the United States. Epoch estimated in September 2026 that they hold 44% of the world's deployed AI compute, with a 90% interval of 23% to 81%. Power, compute and cost are Epoch's estimates from satellite imagery, permits and company statements; four in five IT power estimates fall within a factor of 1.4 of the true value. The owner is the owner of the chips, not necessarily of the building. Capacity after the date of retrieval is Epoch's projection. Coordinates are read from the hub's map page, which publishes them; the downloads do not. Cite Epoch AI, "AI Data Centers". https://epoch.ai/data/ai-data-centers. Retrieved 28 October 2026 via AI Safety Tracker, https://aisafetytracker.org/developers/deepseek/. AI Datacentre Index This site AI datacentre sites that Epoch AI's hub does not hold, recorded from the statements of developers, named tenants and planning authorities. Australia only so far. A site is included only where a primary source says it is built for AI compute or names the AI deployment inside it. Capacity is as the developer states it, IT or facility power as labelled, and is not an estimate; most of it is proposed and not yet approved. Coordinates are this site's placement of each stated address or area. Cite AI Safety Tracker, "AI Datacentre Index". https://aisafetytracker.org/datacentres/#added. Retrieved 28 October 2026.
Energy and emissions
What DeepSeek has published about the energy behind its models, and what permits and press reporting add.
- Training energy or emissions Partly published
- GPU hours only: 2.788 million H800 GPU hours for DeepSeek-V3. No energy or emissions figure. DeepSeek-V3 report
- Where it trains and serves Not disclosed
- Not stated.
- What powers those sites Not disclosed
- Not stated.
- Energy per prompt Not disclosed
- No per-prompt figure.
- - to train DeepSeek-V3 (2024), developer technical report
"Published" means the developer's own statement; "reported by others" means a permit, a regulator or the press. A company's clean-energy claims cover its whole operation, never one model. Environmental Disclosure Index This site Our own reading of what frontier developers and the operators of their data centres have published about training energy, emissions, power sources and energy per prompt, with permits and press reporting marked separately from company statements. Most cells record an absence. Company figures are self-reported and mostly unaudited, and market-based emissions reflect contracts rather than the grid at the hour of use. Cite AI Safety Tracker, "Environmental Disclosure Index". https://aisafetytracker.org/environment/. Retrieved 23 October 2026. Environmental Disclosure Index This site Our own reading of what frontier developers and the operators of their data centres have published about training energy, emissions, power sources and energy per prompt, with permits and press reporting marked separately from company statements. Most cells record an absence. Company figures are self-reported and mostly unaudited, and market-based emissions reflect contracts rather than the grid at the hour of use. Cite AI Safety Tracker, "Environmental Disclosure Index". https://aisafetytracker.org/environment/. Retrieved 23 October 2026.
Usage
The token volumes and user counts DeepSeek has stated, each in its own scope.
No volume or user count found.
Volumes are normalised to a month and are comparable only within one company's own scope. Every series with its scope is on the usage page. AI Usage Disclosure Index This site Our own compilation of the token volumes, user counts and shares of use that companies and studies have put on the record, each linked to where it was stated. Scopes, units and token definitions differ between companies, so no two figures are comparable without the scope printed beside each. Figures known only through press reporting are marked as such. Not a market share estimate, and not a measure of any company's actual volume beyond what it chose to state. Cite AI Safety Tracker, "AI Usage Disclosure Index". https://aisafetytracker.org/usage/. Retrieved 23 October 2026. AI Usage Disclosure Index This site Our own compilation of the token volumes, user counts and shares of use that companies and studies have put on the record, each linked to where it was stated. Scopes, units and token definitions differ between companies, so no two figures are comparable without the scope printed beside each. Figures known only through press reporting are marked as such. Not a market share estimate, and not a measure of any company's actual volume beyond what it chose to state. Cite AI Safety Tracker, "AI Usage Disclosure Index". https://aisafetytracker.org/usage/. Retrieved 23 October 2026.
Not recorded for DeepSeek: a published frontier safety framework; a site in Epoch's data-centre hub; a stated token volume.