Key Takeaways
- DeepSeek founder Liang Wenfeng’s leaked investor meeting reveals AGI research is the primary focus; consumer and enterprise products are described as ‘byproducts’ of the AGI journey.
- The company’s V4-Pro model undercuts GPT-5.5 API pricing by over 90% on output tokens while matching Gemini 3.1 Pro on coding benchmarks.
- DeepSeek plans to at least double every department through a hiring spree, explicitly framing it as pursuit of AGI.
- The technical roadmap progresses from chain-of-thought reasoning to agents, continual learning, and eventually embodied intelligence.
- API pricing is designed to recover hardware costs in roughly 10 months, not to maximize profit, reinforcing the research-first identity.
The DeepSeek Doctrine: AGI Over Everything
A leaked investor meeting transcript from late May 2026 reveals DeepSeek founder Liang Wenfeng’s uncompromising focus on artificial general intelligence over near-term revenue. The Chinese AI lab, known for the open-source R1 reasoning model and recently released V4-Pro, positions consumer and enterprise products as secondary outcomes of its AGI mission. The strategy distinguishes DeepSeek from AI companies building standalone product businesses around their models.
Table of Contents
The AGI Roadmap: From Chain-of-Thought to Embodied Intelligence
According to the transcript circulated internally, Liang Wenfeng outlined a technical progression that starts with chain-of-thought reasoning and moves through AI agents to continual learning, and eventually embodied intelligence. He views continual learning as the next major breakthrough after agents, a step toward systems that can learn from experience without requiring users to provide context afresh each time.
DeepSeek prioritizes coding agents, as these tools have the potential to enhance both its product offerings and internal research workflows. The firm is explicitly not focusing on 3D generation, video creation, or world models, considering them peripheral to its core intelligence mission. Nonetheless, multimodal functionalities will still be developed for use in its consumer and enterprise applications.
The deep research-first culture extends to economics. As reported by TechNode, Liang explained that API pricing is set to recover hardware costs in approximately ten months, rather than maximize profit. This approach keeps costs extraordinarily low: DeepSeek’s V4-Pro charges $0.435 per million input tokens and $0.87 per million output tokens. Comparatively, that is roughly 11.5x cheaper on input and 34x cheaper on output than OpenAI’s GPT-5.5, and 4.6x cheaper on input than Gemini 3.1 Pro. A representative workload of 10 million input plus 2 million output tokens costs around $6 on DeepSeek, versus $44 on Gemini and $110 on GPT-5.5.
As revealed in a leaked investor meeting transcript, Liang Wenfeng characterized open-sourcing as both a deeply held conviction and a deliberate limitation that yields technical advantages. DeepSeek keeps its strongest models open source and uses the same models internally and externally, rather than releasing weaker public versions.
Market Implications: DeepSeek’s Low-Cost Model Shifts Enterprise Calculus
DeepSeek’s approach has already registered measurable market impact, albeit from a small base. As of May 2026, DeepSeek held roughly 4.1% of global web traffic among assistant apps, with a heavily Asian user base and only 1.1% share in the US. By contrast, ChatGPT commanded 46.4% (its first dip below 50%), Gemini 27.7%, and Claude 10.3%.
The pricing gap is staggering. Extrapolating to a SaaS workload of 500 million input and 100 million output tokens per year yields an annual cost of roughly $3,654 on DeepSeek, versus $26,400 on Gemini and $66,000 on GPT-5.5. These figures, based on published API rates from Tech Insider’s analysis, assume no volume discounts and standard pay-as-you-go pricing.
In benchmarks, V4-Pro ties Gemini 3.1 Pro at 80.6% on SWE-Bench Verified, while GPT-5.5 leads at 88.7%. While V4-Pro excels in coding and agentic tasks, it still faces some limitations in pure knowledge retrieval. Check out our detailed analysis of DeepSeek V4’s real-world benchmarks, architecture, and company culture right here. On Terminal-Bench 2.0, an agentic tool-use benchmark, V4-Pro scores 67.9% against GPT-5.5’s 82.7%. DeepSeek’s competitive posture is therefore strongest in cost-sensitive, higher-volume coding and agentic contexts.
To sustain this trajectory, DeepSeek announced a hiring spree on June 26, 2026, aiming to at least double every department, explicitly framed as pursuing AGI. The company’s appeal to talent includes its research-first culture, which may offset stock option offers from more commercially aggressive firms.
Analysts note that DeepSeek’s API pricing, while disruptive, is not yet independently verified as sustainable at production scale for enterprise deployment. The company itself maintains no dedicated sales or customer support team for APIs, consistent with Liang’s characterization of selling APIs as ‘not that attractive.’
The AGI-First Bet: Will Restraint Pay Off?
DeepSeek’s strategy is arguably the purest research-led approach in AI today. By making products and revenue secondary, the company frees itself from short-term profit pressures that could distract from its core AGI mission. The risk is that competitors with deeper product moats and enterprise relationships will capture the market while DeepSeek focuses on a breakthrough that may or may not arrive on its timeline.
If continual learning or the path to embodied intelligence yields an inflection point, DeepSeek’s low-cost, open-source foundation could provide an unparalleled springboard. Until then, the company’s discipline serves as a case study in mission-driven R&D for an industry often tempted by immediate monetization.
For organizations looking to adapt their AI strategies in light of these cost structures and technical roadmaps, Andres SEO Expert offers specialized guidance. Our Programmatic SEO & AI Automation services help integrate AI workflows into content and operations. If you are evaluating hosting infrastructure for AI-heavy applications, our Managed WordPress Cloud Hosting provides optimized environments. Connect with Andres for a tailored consultation, or visit Andres SEO Expert to learn more about our philosophy.
Frequently Asked Questions
What is DeepSeek’s ‘AGI over everything’ strategy?
DeepSeek prioritizes artificial general intelligence research over near-term revenue. Founder Liang Wenfeng views products as secondary outcomes of the AGI mission, distinguishing DeepSeek from companies building standalone product businesses around their models.
How does DeepSeek’s API pricing compare to competitors like OpenAI and Gemini?
DeepSeek’s V4-Pro charges $0.435 per million input tokens and $0.87 per million output tokens. This is roughly 11.5x cheaper on input and 34x cheaper on output than OpenAI’s GPT-5.5, and 4.6x cheaper on input than Gemini 3.1 Pro. A workload of 10M input + 2M output tokens costs ~$6 on DeepSeek vs $44 on Gemini and $110 on GPT-5.5.
Why does DeepSeek open-source its strongest models?
Liang Wenfeng characterizes open-sourcing as a deeply held conviction and a deliberate limitation that yields technical advantages. DeepSeek keeps its strongest models open source and uses the same models internally and externally, rather than releasing weaker public versions.
What is DeepSeek’s technical roadmap to AGI?
DeepSeek’s progression starts with chain-of-thought reasoning, moves through AI agents to continual learning, and eventually embodied intelligence. They prioritize coding agents and consider 3D generation, video creation, and world models as peripheral. Continual learning is seen as the next major breakthrough after agents.
What market share does DeepSeek have compared to other AI assistants?
As of May 2026, DeepSeek held roughly 4.1% of global web traffic among assistant apps, with only 1.1% share in the US. By contrast, ChatGPT commanded 46.4%, Gemini 27.7%, and Claude 10.3%. DeepSeek’s user base is heavily Asian.
What risks does DeepSeek’s research-first approach carry?
The risk is that competitors with deeper product moats and enterprise relationships will capture the market while DeepSeek focuses on an AGI breakthrough that may not arrive on its timeline. Additionally, its API pricing is not independently verified as sustainable at production scale, and the company lacks dedicated sales or customer support for APIs.
