Jan 2025, model, DeepSeek

DeepSeek-R1

A Chinese reasoning model that DeepSeek said performed on par with OpenAI's o1, released with its weights under the MIT licence; a week later Nvidia lost close to $600 billion in market value in a day, and both the export-controls and open-models camps claimed it proved their case.

What it was

On 20 January 2025 the Chinese lab DeepSeek released DeepSeek-R1 and DeepSeek-R1-Zero, reasoning models built on its DeepSeek-V3 base model, with 671 billion total parameters of which 37 billion are active for any given token. DeepSeek's release note claimed "Performance on par with OpenAI-o1" on maths, code and reasoning and put the code and weights under the MIT licence, which permits commercial use and distillation. It also released six smaller models distilled from R1 and built on Meta's Llama and Alibaba's Qwen.

Unlike OpenAI, DeepSeek published how it trained the model. R1-Zero was "trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step," and reasoning behaviours emerged from the reward signal alone. The December 2024 technical report on the V3 base model said it had needed "only 2.788M H800 GPU hours for its full training," using a chip that, Amodei later noted, had been allowed for sale to China under the first US export controls of 2022 and was banned in October 2023.

Read the original

What it changed

On 27 January 2025 Nvidia's shares fell 17 percent, a loss of close to $600 billion in market value and, CNBC reported, the biggest one-day drop ever for a US company. An analyst quoted by CNBC said the release had caused "great angst as to the impact for compute demand, and therefore, fears of peak spending on GPUs."

Within days the leading figures in the export-controls and open-models debates each read the release as support for their position. Dario Amodei wrote in January 2025 that DeepSeek's results "make export control policies even more existentially important than they were a week ago," and that "DeepSeek produced a model close to the performance of US models 7-10 months older, for a good deal less cost (but not anywhere near the ratios people have suggested)." Yann LeCun wrote that people who saw "China is surpassing the U.S. in AI" were "reading this wrong. The correct reading is: 'Open source models are surpassing proprietary ones'," in a LinkedIn post reported by CNBC. Andrew Ng wrote on 29 January that "If the U.S. continues to stymie open source, China will come to dominate this part of the supply chain." On 31 January Sam Altman said in a Reddit question-and-answer session that OpenAI had been "on the wrong side of history" on open source, and that he personally thought OpenAI needed to "figure out a different open source strategy."

The arguments it moved

Who should have access to powerful models?

Ng used R1 to recast the open-weights argument as a question about competition with China. His 29 January 2025 letter said that "Open weight models are commoditizing the foundation-model layer" and that if the United States restricted them, China would "dominate this part of the supply chain." LeCun argued that the lesson was about openness, not nationality: DeepSeek "profited from open research and open source." Critics of open release, including Bengio and Hinton, have argued that weights cannot be recalled once published; R1's weights were published in China, where US rules on model release would not have reached them.

Compare every leader's position on this

Who should set the rules for AI?

Amodei made R1 the centre of his case for chip export controls. His January 2025 essay argued that DeepSeek had trained on a "mix of H100's, H800's, and H20's," that its cost savings were "an expected point on an ongoing cost reduction curve," and that "Well-enforced export controls are the only thing that can prevent China from getting millions of chips, and are therefore the most important determinant of whether we end up in a unipolar or bipolar world." He also wrote that "The performance of DeepSeek does not mean the export controls failed."

Compare every leader's position on this

What can AI learn through experience?

DeepSeek reported that R1-Zero's reasoning ability emerged from reinforcement learning without a supervised warm-up, and published the method. David Silver and Richard Sutton's 2025 essay "Welcome to the Era of Experience" quoted DeepSeek's report that "rather than explicitly teaching the model on how to solve a problem, we simply provide it with the right incentives, and it autonomously develops advanced problem-solving strategies," as evidence that self-generated data could replace data written by people.

Compare every leader's position on this

Positions it bears on

Dario Amodei, Compute and export controls

Because capability tracks compute, he treats denying China advanced chips as the single most consequential policy decision of the decade.

Amodei's January 2025 essay "On DeepSeek and Export Controls" argued that R1 made chip controls more important, not less.

Critics and counterpoints
Andrew Ng, Open versus closed models

Open-weight models are a key part of the AI supply chain, drive down prices, and make systems more secure. He has argued that restricting them would hand that layer of the industry to China.

Ng's 29 January 2025 letter cited R1 as evidence that restricting open models would cede the supply chain to China.

Critics and counterpoints
Yann LeCun, Open vs closed models

Insists that foundation models must be open source so that no handful of companies controls the information people receive, and sees the concentration of AI in proprietary systems as a greater danger than the technology itself.

LeCun read R1 as a win for open-source models over proprietary ones, not for China over the United States.

Critics and counterpoints
Sam Altman, Open versus closed models

He has conceded that OpenAI's closed approach put it on the wrong side of history and has released open-weight models, while keeping frontier models proprietary.

Eleven days after R1's release, Altman said OpenAI had been on the wrong side of history on open source.

Critics and counterpoints
David Silver, Limits of learning from human data

Argues that imitating human data can reproduce human competence but not exceed it, and that in mathematics, coding and science the useful human data has largely been consumed.

Silver's 2025 essay quoted DeepSeek's R1 report as evidence that models can learn reasoning from incentives rather than human examples.

Critics and counterpoints
Jensen Huang, Export controls and China

Huang argues that U.S. restrictions on AI chip sales to China have failed, cost American companies a large market and accelerated Chinese chipmakers, and that the United States wins by getting the world to build on American technology.

R1 wiped close to $600 billion off Nvidia's value in a day; four months later Huang called the export controls meant to hold China back "a failure."

Critics and counterpoints

Sources

Last verified 2026-09-22.