What happened to Deepseek?
The Chinese AI challenger has gone radio silent after promising a GPT-4 rival, raising questions about its future.
DeepSeek, once hailed as a formidable Chinese open-source contender against giants like OpenAI and Anthropic, has entered a concerning period of radio silence. After generating significant excitement in late 2023 and early 2024 with its DeepSeek-V2 series—a 236B parameter Mixture-of-Experts model that offered strong performance at a lower cost—the company's promised follow-up, a 'V4' model intended to compete with GPT-4, has failed to materialize. Its last major communication was months ago, and its GitHub activity has slowed to a trickle, a stark contrast to the frenetic pace of releases from Western labs like Meta, which has shipped multiple Llama 3 variants.
This disappearance has sparked intense speculation within the AI community. The dominant theory points to a severe funding crunch, as the astronomical compute costs for training frontier models collide with the lack of a clear monetization path for purely open-source labs. Other possibilities include a strategic retreat to develop a more proprietary, closed model for commercial launch, or significant technical hurdles encountered during the V4 training cycle. The silence is particularly poignant as it highlights the fragility of the open-source AI ecosystem, where even well-regarded teams can vanish, leaving projects and dependent developers in limbo.
The community's concern is amplified by the context of Meta's 'open' but not fully open-source approach with Llama 3, which uses a restrictive license. DeepSeek's potential fade weakens the field of truly permissive, high-performance open models, potentially consolidating power with a few well-funded corporations. The unanswered question remains: is DeepSeek gearing up for a surprise comeback, or has it become a cautionary tale about the unsustainable economics of open-source AI research at the frontier?
- DeepSeek's last major model was the 236B parameter DeepSeek-V2, released months ago with no 'V4' successor in sight.
- The company's public communication and GitHub activity have nearly halted, contrasting with Meta's aggressive Llama 3 release schedule.
- The primary speculation centers on a funding crisis, as training costs soar without a clear revenue model for open-source AI labs.
Why It Matters
The fate of DeepSeek is a stress test for the sustainability of open-source, frontier AI development outside major tech conglomerates.