Today marked a significant day for Together AI as it hosted its first-ever AI Native Conference. The event showcased some of the latest breakthroughs in AI infrastructure, open source research, and reinforcement learning. It brought together builders, researchers, and industry leaders to share insights on creating the next generation of AI applications.

Rapid Growth and Industry Impact

Together AI has seen impressive growth recently. The company now serves thousands of customers and supports over one million developers worldwide. Its annual contract revenue has grown tenfold compared to last year. Among its major deals are 27 contracts exceeding one million dollars and a single deal surpassing one billion dollars.

The company has become a key infrastructure provider for leading AI-native companies like Cursor, Decagon, and Cartesia. It helps power their AI systems with production-scale inference, pre-training, and model shaping. This positions Together AI as a crucial player in the AI ecosystem, bridging cutting-edge research and practical deployment.

Innovations and Research Breakthroughs

During the conference, Together AI announced several new advancements that push the boundaries of AI technology. These include improvements in kernels, reinforcement learning, and inference optimization. The company’s focus remains on making AI training and inference faster and more efficient, helping businesses of all sizes leverage generative and agentic AI.

One of the key innovations is FlashAttention 4, the latest version of a popular kernel that powers most large language models in production. It offers up to four times better performance at long sequence lengths, which is crucial for tasks like coding and document understanding. This reduces the gap between theoretical speed and real-world performance for long-context workloads.

Another major announcement is a new Reinforcement Learning API that separates inference from training. This allows companies to run distributed reinforcement learning pipelines across multiple locations, something that was previously only possible for organizations with huge GPU clusters. Additionally, ThunderAgent, an open-source system, now provides up to 3.6 times more throughput and less memory use for serving and training AI agents.

Furthermore, ATLAS-2 was introduced as a system that uses real-time user data to adapt and optimize AI models instantly. It can deliver inference results 1.5 times faster, helping applications respond more quickly to user needs. These advancements highlight Together AI’s commitment to accelerating AI development and deployment across industries.

Fostering the AI-Native Community

The AI Native Conf was created to gather the next generation of AI builders. It provided a platform for sharing innovations, discussing challenges, and forging new partnerships. The event emphasized the importance of collaboration between research labs and production systems, a core strength of Together AI.

Vipul Ved Prakash, co-founder and CEO, spoke about how AI is evolving faster than any previous technological shift. He emphasized that the companies being built today are fundamentally different, with a focus on running AI at scale in real-world environments. He highlighted the company’s unique position, where the same researchers developing foundational AI work are also shipping it into production systems used by customers.

Overall, the event underscored Together AI’s role at the forefront of AI innovation. By combining industry-leading research with practical solutions, the company is helping shape the future of AI technology on a global scale.

Inspired by

• https://ai-techpark.com/together-ai-marks-key-milestones-at-ai-native-event/