DeepSeek is a Chinese AI lab that shook the world by releasing frontier AI as 'open weights.' In July 2023, Liang Wenfeng, who leads the quant hedge fund High-Flyer, founded it (he holds about 84%), and it has released the weights and methods of all its models—Coder, V2, V3, R1—in a form anyone can download, modify and build on. This contrasts with OpenAI's and Anthropic's leading models being closed. R1, in January 2025, achieved reasoning performance rivaling OpenAI's o1 at 1/10 to 1/100 the cost (V3's final training was about $5.6M), done under US chip-export controls with H800s and MoE (mixture of experts). DeepSeek's assistant overtook ChatGPT to top the US App Store's free chart, and Nvidia lost $600B in a single day (the largest ever)—dubbed the US AI world's 'Sputnik moment.' V2's price war moved China's AI market toward open source, and it has stated it has no near-term commercialization plans.
●○○ low
The assessment is on hold, awaiting a build-up of confirmed evidence.= non-additive meter
DeepSeek: Frontier AI released as open weights—net value is contested (on hold). At present independent evidence is scarce, so the assessment is on hold. Unconfirmed concerns are placed under “Watching.” (As of 2026-Q3; estimate based on public information.)
Main narrative
One person’s story (N1)
+ A single story
A +N1 (one person’s story) will be added once an independent source is confirmed.
Positive / negative effects
+ effects
- See the N1 above for the main positive story; independently verified + will be added over time.
− effects (confirmed)
- No confirmed −.
- C
- h
- i
- n
- a
- '
- s
- c
- o
- n
- t
- e
- n
- t
- r
- e
- g
- u
- l
- a
- t
- i
- o
- n
- (
- c
- e
- n
- s
- o
- r
- s
- h
- i
- p
- )
- ;
- d
- a
- t
- a
- j
- u
- r
- i
- s
- d
- i
- c
- t
- i
- o
- n
- /
- p
- r
- i
- v
- a
- c
- y
- ;
- d
- u
- a
- l
- u
- s
- e
- a
- n
- d
- d
- i
- s
- t
- i
- l
- l
- a
- t
- i
- o
- n
- s
- u
- s
- p
- i
- c
- i
- o
- n
- s
- ;
- s
- a
- f
- e
- t
- y
- i
- s
- s
- u
- e
- s
- o
- f
- A
- G
- I
- -
- o
- r
- i
- e
- n
- t
- e
- d
- f
- r
- o
- n
- t
- i
- e
- r
- A
- I
- ;
- t
- h
- e
- n
- e
- t
- e
- f
- f
- e
- c
- t
- o
- f
- o
- p
- e
- n
- n
- e
- s
- s
- i
- s
- c
- o
- n
- t
- e
- s
- t
- e
- d
- .
- The real picture of downstream innovation/access from open weights; the handling of content regulation and data jurisdiction; its stance on AI safety; long-term assessment of impact.
A second look
Releasing open weights opens access to cutting-edge AI to anyone (including researchers, developers, and resource-poor institutions/countries) and prevents AI power from concentrating in a handful of companies—a genuine plus as a non-financial value, a public good. But at the same time DeepSeek is a frontier AI lab pursuing AGI, entangled with China's content regulation (censorship of politically sensitive topics), data jurisdiction, dual use, distillation suspicions, and the safety issues of frontier AI development itself. Net value to those protected (People, future generations) is at present undetermined and contested, and, founded in 2023, its impact is still being assessed. Acknowledging the plus of openness, a definitive A–E / out-of-scope judgment is premature, and it is on hold.
Sources
How to read this assessment
- Reachable upper bound (ceiling): a confirmed − sets the ceiling, and independently verified + decide the position within it. + do not cancel out −.
- The weight of evidence is not symmetric: only confirmed − are counted; the volume of disputes or allegations goes under “Watching.” + are counted from independent evidence, while an organization’s own PR is treated as “reference.”
- Size is not value: scale is not used in the assessment. Matters that stay within money or competition—investors, shareholders, sanctions, trade secrets—are also excluded.
- The letter (assessment) and certainty (how reliable the information is) are separate axes.
This is a translation; the Japanese version is authoritative. The assessments here are generated automatically by AI based on published criteria. The operator does not alter individual results. Because they are AI-generated they may contain errors, and they are opinion and commentary, not statements of fact. Where evidence is insufficient, the entry is marked “On hold.” Requests for correction are accepted via the form.
Terms: Narrative Value = an assessment (A–G) of the distance between the story an organization tells and its reality / Ceiling meter = a visualization of the reachable upper bound / Watching = unconfirmed matters not counted / Protected stakeholders = people, animals, nature, and future generations. | Generated by: AI | As of: 2026-Q3 | Back to top