ClaimsDario Amodei

“US companies will soon follow suit — and they won’t do this by copying DeepSeek, but because they too are achieving the usual trend in cost reduction.”

Dario Amodei, essay author 13 October 2022 On DeepSeek and Export Controls (darioamodei.com)

Prediction Economics and markets

US AI companies will soon achieve similar cost reductions, through the usual cost-reduction trend rather than by copying DeepSeek.

Unresolvable The public record could not settle it.

Prediction: Statements about events outside the person's control, checked once they fell due: did they come true?

The second judge would not resolve this claim; the verdict shown is the first judge's.

What would count as true or false

The reference to DeepSeek places this statement in the context of Amodei's January 2025 essay on DeepSeek. Read 'soon' as within about 12 months of that essay, so by January 2026. TRUE: by then, at least one major US AI company (for example OpenAI, Anthropic, Google or Meta) has released models matching DeepSeek-V3/R1-level capability at comparable or lower cost (training cost or per-token price), credibly attributed to the company's own efficiency progress. There must be no evidence that the gains came mainly from adopting DeepSeek's specific methods. MOSTLY TRUE: US companies reach comparable cost-efficiency in the window, but slightly later than 'soon' implies (within about 15 months), or the parity is somewhat less close. MIXED: some US companies reach comparable price and performance, but credible reporting shows they partly adopted DeepSeek-specific techniques, or parity holds on some measures only. MOSTLY FALSE: cost reductions happened but stayed clearly short of DeepSeek's price and performance in the window, or were mainly attributed to copying DeepSeek. FALSE: there is no meaningful narrowing of the cost gap by US companies in the window.

Set before the research began.

Evidence (7)

  1. E1 Anthropic's CEO says DeepSeek shows US export rules are working | TechCrunch TechCrunch · 29 January 2025

    ““U.S. companies [are also] achieving the usual trend in cost reduction,” Amodei added.”

  2. E2 Anthropic's CEO says DeepSeek shows US export rules are working | TechCrunch TechCrunch · 29 January 2025

    “[This is] an expected point on an ongoing cost reduction curve. What’s different this time is that the company that was first to demonstrate the expected cost reductions was Chinese.”

  3. E3 Anthropic's CEO says DeepSeek shows US export rules are working | TechCrunch TechCrunch · 29 January 2025

    “The efficiency innovations DeepSeek developed will soon be applied by both U.S. and Chinese labs to train multi-billion dollar models.”

  4. E4 DeepSeek R1 0528 (May '25) vs o3-mini: Model Comparison | Artificial Analysis Artificial Analysis · 1 January 2026

    “o3-mini is cheaper. At a 7:2:1 cache hit/input/output ratio, o3-mini costs $1.05 per 1M tokens (based on the median across providers serving the model), compared with DeepSeek R1 0528 (May '25) at $1.82 per 1M tokens (based on the median across providers serving the model).”

  5. E5 raw AI Wiki (aiwiki.ai)

    “although DeepSeek's official API at the time offered the same model at lower rates (approximately $0.55 input and $2.19 output per million tokens)”

  6. E6 raw AI Wiki (aiwiki.ai)

    “OpenAI's pricing decision was widely interpreted in trade press as a competitive response to DeepSeek's January 20 release.”

  7. E7 Gemini 2.0 Flash vs DeepSeek V3 ModelRefs · 26 December 2024

    “Input price | $0.27/M tokens | $0.1/M tokens”

Reasoning

The evidence shows some price comparisons: o3-mini has a lower blended median price than DeepSeek R1 0528 [E4], DeepSeek's official API had lower rates [E5], and an unlabelled price row exists [E7]. None of the evidence establishes that a US model matched DeepSeek-V3/R1 capability or why its costs fell, so the criteria cannot be applied.

Cite this claim

Dario Amodei, “US AI companies will soon achieve similar cost reductions, through the usual cost-reduction trend rather than by copying DeepSeek.” (said 13 October 2022). Verdict: unresolvable. Reliability API, claim c_864124afd03f, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_864124afd03f

This claim is in dataset release v1.1.2, with the same verdict.

Reply to this claim

For Dario Amodei, or someone speaking for them: a correction or counter-evidence. Replies are read before anything is shown; an approved reply shows your name, role, date, statement and link. Your email address is never shown and is deleted 30 days after the decision (privacy).

More prediction claims by Dario Amodei

  1. Resolved false Prediction 13 October 2022
    “This will rapidly cease to be true as everyone moves further up the scaling curve on these models.”

    The situation where several companies can produce good reasoning models will rapidly cease as everyone moves further up the scaling curve.

  2. Mostly false Prediction 23 September 2026
    “We believe AI for biology is on a similar exponential trend.”

    Amodei said that "we" believe AI for biology is on an exponential trend similar to the one he described for AI in mathematics.

  3. Mostly true Prediction 28 February 2026
    “We're gonna be fine. The-- the impact of this designation is fairly small.”

    Anthropic will be fine, and the impact of the supply chain risk designation on it will be fairly small.

  4. Resolved false Prediction 10 March 2025
    “where AI is writing 90 percent of the code”

    Within three to six months of March 2025, AI will be writing 90 percent of code.

  5. Resolved false Prediction 10 March 2025
    “And then in twelve months, we may be in a world where AI is writing essentially all of the code”

    Within twelve months of March 2025, AI may be writing essentially all of the code.

All of them

Verdicts are made automatically; how verdicts are made.