Together AI's Open-Model Advantage

Together used open models to turn raw GPU access into a faster product loop for startups. Instead of asking teams to first pick hardware, reserve capacity, and wire up model serving, it let them try many ready to run models, fine tune them on their own data, and ship through one API. That made Together closer to an AI application layer, while CoreWeave and Lambda were earlier centered on selling compute itself.1 2 3 4

• The practical wedge was model choice. Together offered 100 plus plug and play open models, from Llama to Mistral, so a team building support, search, or coding features could swap models quickly instead of committing to a single closed provider or managing open source stacks themselves.1 2
• The business model matched that workflow. Together sold usage priced APIs and managed training on top of rented GPU capacity, while CoreWeave and Lambda were built more around hourly infrastructure and larger reservations. That fit startups with bursty demand who wanted to avoid paying for idle GPUs.1 2 3 4
• That positioning also explains why Together scaled differently. In April 2024, annualized revenue was estimated at $44M versus CoreWeave at $465M in 2023 and Lambda at $250M in 2023, because Together captured a thinner software and resale layer while owning more of the developer relationship.1 2 3 4

The next phase is a race to keep the open model layer valuable as base compute and model hosting get crowded. The winners will be the platforms that make open models easiest to test, tune, route, and run in production, not just the ones with the most GPUs.2 4