Together AI: Open-Source LLM Training & Inference Infrastructure
Together AI provides cost-efficient infrastructure for training and deploying open-source language models. Integrates with PyTorch and Hugging Face. Founded 2022 in San Francisco.
Together AI is headquartered in San Francisco with compute infrastructure distributed globally enabling local training and deployment. The company serves international customers through regional compute availability.
Founded: 2022 · HQ: San Francisco, California, USA · Team: 50-100 · CEO: Isaac Ong · Funding: $50+ million · Valuation: $600+ million (estimated)
About Together AI
Together AI was founded in 2022 by Isaac Ong, Tim Tully, Ramesh Chandra, and others from premier AI research institutions and technology companies. The company is headquartered in San Francisco with approximately 50-100 employees focusing on infrastructure, research, and operations. Together AI identified critical gaps in accessible infrastructure for training and deploying language models, creating platform solutions enabling researchers and organizations to participate in model development without massive capital requirements typical of incumbent approaches. Together AI's flagship offering is a distributed infrastructure platform enabling efficient training of large language models across distributed GPU clusters. The platform abstracts complexity of multi-machine training, enabling researchers and organizations to train models on proprietary or specialized data efficiently. Together Compute provides cost-optimized inference infrastructure for serving trained models at scale. The company released open-source models trained on the Together platform, demonstrating platform capabilities and supporting open-source AI community. Together designed integrations with popular frameworks (PyTorch, Hugging Face) enabling seamless adoption by existing development workflows. The company has secured significant funding supporting platform development and operational scaling. Together AI raised Series A and subsequent funding rounds totaling approximately $50+ million from prominent venture investors recognizing the strategic importance of accessible model training infrastructure. The funding has enabled expansion of computing capacity, team growth, and development of increasingly sophisticated platform features. Investors believe alternative infrastructure providers are essential for competitive AI development landscape preventing monopolistic control by dominant computing providers. Together AI's research contributions focus on training efficiency, distributed systems, and open-source model development. The company publishes research on optimization techniques, training algorithms, and benchmark results from open-source models trained on the platform. Together participates in community efforts advancing open-source AI and maintains partnerships with research institutions. The team's expertise in distributed systems and machine learning enables technical innovation in platform capabilities. Recent developments include releasing open-source models trained on Together infrastructure (RedPajama, OpenChatKit), expanding compute capacity supporting larger-scale training, and launching new features enabling fine-tuning and adaptation of models. The company announced improved cost efficiency for inference through optimization techniques and hardware partnerships. Together began offering managed model deployment services for organizations lacking internal infrastructure expertise. The company expanded globally with compute availability across regions serving international customers. Competitively, Together AI differentiates through accessibility, cost-efficiency, and commitment to open-source development. While large technology companies control proprietary training infrastructure, Together democratizes access enabling smaller organizations and research groups to train models competitively. The company's cost-focused approach appeals to budget-conscious organizations. Emphasis on open-source community aligns with diverse stakeholders preferring distributed AI development. Integration with Hugging Face ecosystem enables broad adoption by developers and researchers using standard tools. Together AI operates with commitment to enabling diverse participation in AI development. The company maintains transparent pricing, publishes research advancing the community, and actively supports open-source initiatives. The company believes accessible infrastructure is essential for competitive, innovative AI development landscape. Documentation emphasizes responsible training practices and ethical AI development. Together engages with stakeholders on governance and ownership structures for AI systems. The company's founding philosophy emphasizes that artificial intelligence development should be distributed, accessible, and community-driven rather than concentrated in hands of dominant corporations. Together believes that democratizing access to training infrastructure enables innovation, competition, and beneficial outcomes across diverse organizations and communities. This vision guides platform development, investment decisions, and community engagement.
Mission
To democratize access to AI infrastructure enabling diverse organizations and researchers to train and deploy language models efficiently and cost-effectively.
Products
- Together Compute
- Inference Engine
- API Service
- Integration Ecosystems
Compliance
SOC 2 Type II
Links
Website · GitHub · Twitter · LinkedIn · Blog · Docs
Frequently Asked Questions
What is Together AI?
Together AI is a San Francisco company founded in 2022 that provides cloud infrastructure for training, fine-tuning, and running open-source AI models. It offers fast, cost-efficient inference and GPU clusters as an alternative to closed-model APIs. Developers use it to deploy open models such as Llama, Mixtral, and DeepSeek at scale.
What does Together AI offer?
Together AI provides a serverless inference platform, fine-tuning services, and dedicated GPU clusters optimized for open models. Its inference engine is engineered for high throughput and low cost using optimized kernels and techniques like speculative decoding. It also contributes open research, including the RedPajama datasets.
Who is Together AI for?
Together AI targets developers, startups, and enterprises that want to build on open-source models rather than depend on closed APIs from OpenAI or Anthropic. It suits teams needing custom fine-tuned models, data control, or lower inference costs at scale. Researchers also use it for large training runs.
How is Together AI different from OpenAI?
OpenAI serves its own closed models through a single API, while Together AI hosts a broad catalog of open-source models you can run, fine-tune, and own. This gives customers more control over cost, data, and model choice. Together emphasizes open infrastructure rather than proprietary models.
How much does Together AI cost?
Together AI charges per-token for serverless inference and hourly for dedicated GPU instances, with pricing that varies by model and hardware. Open-model token rates are typically far lower than closed-model APIs. Exact rates are published on its pricing page and change over time.
What is RedPajama?
RedPajama is an open dataset and model initiative co-led by Together AI to reproduce and openly release large training datasets for language models. It reflects the company's commitment to open AI research. The datasets have been widely used to train community models.
How is Together AI funded?
Together AI is headquartered in San Francisco and has raised several large venture rounds, reaching a multibillion-dollar valuation by 2025 with backing from investors including NVIDIA and General Catalyst. The funding supports the expansion of its GPU cloud and inference platform.
Is Together AI open source?
Together AI is built around open-source models and contributes open datasets and research, though its hosting platform itself is a commercial service. Customers run open-weight models with full control over fine-tuning and deployment. This open-first stance is central to its positioning.