DeepSeek founder vows to keep top AI models open-source

Liang Wenfeng, the founder of DeepSeek, laid out an expansive vision for his company's future in a nearly four-hour investor meeting, declaring that the Chinese AI laboratory's most powerful models — including upcoming releases — will remain open-source and that closed-source development offers no meaningful advantage.

The contents of the session, which took place as part of DeepSeek's first-ever external funding round, were published this week by Tencent Technology in a 118-entry transcript that offers the most detailed public account yet of how Liang intends to steer the Hangzhou-based company.富途資訊+1

Open-Source as Strategy, Not Sacrifice

Liang was direct in dismissing the notion that open-source development comes at a commercial cost. "I don't see any clear benefits to keeping them closed-source — no inherent advantages whatsoever," he said, according to the transcript. He pointed to ByteDance's closed-source approach as a counterexample that has yielded no obvious strategic benefit.FuTu News

He confirmed that the model DeepSeek releases publicly is identical to its internal deployment. "We would never release an inferior open-source model while using a better one internally — that simply doesn't happen. They are identical," Liang said.FuTu News

The company's forthcoming V4 model, which will feature native multimodal support, is expected to follow the same open-source path. Liang argued that even with full transparency, the barriers to effective deployment remain high enough that open-sourcing creates goodwill without enabling meaningful competition. "Open-sourcing will not impact our revenue," he stated. "I believe it has absolutely no effect on our business model."ChinaBiz Insider+1

A $7 Billion War Chest for AGI

The investor session came after DeepSeek closed a funding round exceeding 50 billion yuan (approximately $6.9 billion), valuing the company at roughly $51 billion pre-money. The investor roster includes Tencent at 10 billion yuan, CATL Contemporary Amperex Technology Co., Limited at 5 billion yuan, and NetEase ↘ 0.02%, JD.com ↗ 0.15%, and IDG Capital at 3 billion yuan each. Liang himself contributed the largest single check at 20 billion yuan.Forbes+2

Liang framed the fundraise not as a commercialization push but as a retention tool, saying team stability is DeepSeek's "only non-negotiable core interest." He outlined a four-stage AGI roadmap — chain-of-thought reasoning (completed in 2025), agent frameworks (2026's focus), continuous learning, and eventually self-iterating intelligence — while explicitly ruling out video generation and 3D modeling as distractions from the "intelligence mainline."ChinaBiz Insider+1

The Case Against Monopoly

Throughout the session, Liang returned to a philosophy of deliberate restraint. He argued that the AI opportunity is so vast that attempts to monopolize its value are self-defeating. "Those who aim to take more will be defeated by those who aim to take less," he said, adding that DeepSeek's API pricing targets a ten-month hardware payback cycle rather than maximum profit extraction.ChinaBiz Insider+1

Liang projected that large language models would support only a handful of frontier providers globally and that none would earn outsized profits. "Those with better cost control will earn slightly more; those with poor cost control will earn less — that's all there is to it," he said.FuTu News