ByteDance Seed
Foundation Model
Campus Recruitment
New graduate recruitment: for undergraduates, Master's and PhD students graduating between September 2026–August 2027
Internship recruitment: for students graduating in September 2027 or later
Seed Foundation Model Campus Recruitment is a ByteDance Seed recruitment initiative for campus talent. We firmly believe that true technological progress stems from continuously seeking to scale new heights. Amid the immense opportunities of the AI era, the Seed team does not stop at model iteration; instead, we dive into core cutting-edge technologies, drive breakthroughs in the next-generation AI paradigm, and constantly advance the boundaries and frontiers of intelligence. Joining now, you will be at the forefront of building cutting-edge AI capabilities and driving the next breakthrough.
Who we are looking for
Strong faith in technology and visionary goals
Commitment to advancing the frontiers of intelligence; ability to tackle the industry's toughest challenges with perseverance.
Excellent research and practical capabilities
Deep technical expertise and proven work in specific foundation model domains; relentless spirit for research and strong hands-on capabilities.
Strong curiosity and motivation
Curiosity about the unknown; focus on the essence; motivation to go above and beyond in solving problems.
Open-minded and willing to collaborate
Scaling is a systemic engineering feat; we expect you to collaborate seamlessly with the team in the R&D of multimodal foundation models.
Strong faith in technology and visionary goals
Commitment to advancing the frontiers of intelligence; ability to tackle the industry's toughest challenges with perseverance.
Excellent research and practical capabilities
Deep technical expertise and proven work in specific foundation model domains; relentless spirit for research and strong hands-on capabilities.
Strong curiosity and motivation
Curiosity about the unknown; focus on the essence; motivation to go above and beyond in solving problems.
Open-minded and willing to collaborate
Scaling is a systemic engineering feat; we expect you to collaborate seamlessly with the team in the R&D of multimodal foundation models.
Strong faith in technology and visionary goals
Commitment to advancing the frontiers of intelligence; ability to tackle the industry's toughest challenges with perseverance.
Excellent research and practical capabilities
Deep technical expertise and proven work in specific foundation model domains; relentless spirit for research and strong hands-on capabilities.
Strong curiosity and motivation
Curiosity about the unknown; focus on the essence; motivation to go above and beyond in solving problems.
Open-minded and willing to collaborate
Scaling is a systemic engineering feat; we expect you to collaborate seamlessly with the team in the R&D of multimodal foundation models.
Technical research directions
Foundation models
- Structural optimization and training paradigms for efficient foundation models
- The science of data compression in language models
- Code agent RL & test-time scaling
- Long-horizon tasks
- Multi-agent reinforcement learning (MARL)
- RL scaling
- AI4Math: Frontier mathematical reasoning and discovery
- Real-time interaction for foundation models
- More advanced model distillation techniques
- Next Paradigm (Edge) Exploration: Intrinsic Rewards and Agency
- Memory and reasoning, self-supervision, fusion, and unified modeling tools
Visual intelligence
- Multimodal foundation data pipelines
- Scientific and efficient evaluation frameworks for video and image models
- New paradigms in multimodal pre-training
- General and efficient visual representation learning
- Unified visual reasoning models for generation and understanding
- Multimodal agents for complex, long-horizon tasks
- Tool use and vision-based action
Speech intelligence
- Unified generation and understanding in multimodal interaction, and multimodal joint learning
- Memory and learning mechanisms in personalized dialogue, as well as online learning
- RL algorithms and system optimization for speech/audio multimodal tasks, and speech agents
- Comprehensive audio understanding, audio generation, and music generation
Machine learning systems
- Enhance training stability and MFU under ultra-large-scale clusters
- Design end-to-end RL systems for foundation models to support complex real-world environments
- Work on efficiency challenges in large model inference systems, including multi-node inference, parallel inference strategies, and scheduling optimization.
- Explore more advanced model structures, training patterns, and inference paradigms based on next-gen hardware and unified generation-understanding architectures
- Under new hardware architectures, enhance compilers for high-performing operators and jointly optimize computation and communication
LLM applications
- Revolutionize search content understanding based on LLM
- Design general-purpose search solutions powered by foundation models
- Train search agents
- Improve model accuracy across common tasks
- Design model memory systems
- Design personalization schemes for models
- Model personality settings and linguistic styles
- Apply user feedback signals
Embodied intelligence
- General-purpose robotic foundation models
- Robot data algorithms
- Robot reinforcement learning
What you can achieve here
Work on the most challenging frontiers in AI
Participate in building industry-leading Seed foundation models and GenMedia models, and explore the next generation of AI.
Full support in accessing R&D resources
With our robust AI infra, you will have access to sufficient computing power, data, and engineering resources to focus on overcoming core technical challenges.
Create real value for the world
We have competitive advantages in both consumer and enterprise products. Your technical achievements will deliver tangible value to hundreds of millions of users.
Get rewards matching technical contributions
We offer highly competitive compensation packages. Eligible new graduates and top-performing interns will have the chance to join an extra incentive program designed for AI talent.
Join us to make industry-leading achievementsJoin us to make industry-leading achievementsJoin us to make industry-leading achievements
Seed2.0
With fully upgraded multimodal understanding and significantly enhanced LLM and Agent capabilities, it can tackle real-world long-chain complex tasks, with overall performance ranking among the industry's top tier.
Seedance 2.0
Adopting a unified multimodal audio-video joint generation architecture, it integrates comprehensive multimodal content reference and editing capabilities, achieving SOTA-level usability in complex scenario generation.
Seedream 5.0 Lite
As a further exploration toward a unified multimodal model, it comprehensively improves understanding, reasoning, and generation, with deep thinking and real-time retrieval enhancement capabilities.
Seed Realtime Voice
The end-to-end integration of speech and text modalities continuously delivers a voice interaction experience with extremely low latency and human-like conversation quality to over 100 million users.
Seed GR-RL
A robot learning framework for long-horizon dexterous manipulation tasks that overcomes fine-manipulation bottlenecks in VLA and achieves real-world reinforcement learning shoelace threading for the first time.
Q&A section
What is the relationship between Seed Foundation Model Campus Recruitment and the previous Top Seed talent program?
The Seed-related recruitment for 2027 new graduates and student interns focusing on foundation model development has been consolidated under the Seed Foundation Model Campus Recruitment program this year. The Top Seed talent program will no longer operate as a separate initiative. As Seed continues to advance in foundation model technology, we are tackling the most challenging work in building the next generation of AI. We seek talent with strong technical insights and innovation capabilities to advance the frontiers of intelligence. "Lead people based on their qualities" has long been a core talent philosophy at ByteDance. We remain committed to our principle of "prioritizing talent potential over experience", providing top talent with a solid foundation and unlimited opportunities to grow.
As a Class of 2027 candidate who has already received an internship offer from another team, can I still apply for full-time new graduate positions in the Seed Foundation Model Campus Recruitment program?
Each candidate may only participate in one job application process at a time. For special circumstances, please contact the recruiting HR for your current process for further assistance.
What is the interview process of Seed Foundation Model Campus Recruitment for the Class of 2027 graduates and interns?
Usually, the interview process consists of 3–5 rounds of technical interviews and 1 HR interview. Specific arrangements may be adjusted based on the role and candidate situation, and the final process is subject to HR's actual arrangement.
