Together AI raises reserved H100 pricing while expanding its model catalog
By Rob Litterst, Co-founder, PricingSaaS
Published
Together’s pricing page · Change detected
What changed
Together AI raised reserved-cluster pricing for NVIDIA HGX H100 GPUs across every commitment tier: 31-90 days moved from $3.59 to $3.69/GPU/hr, 91-180 days from $3.29 to $3.45/GPU/hr, and 181+ days from $3.09 to $3.19/GPU/hr. On-demand H100 pricing held flat at $3.99/hr. Separately, Together added two serverless inference models to its catalog: Kimi K3 at $3.00/$15.00 per 1M input/output tokens ($0.30 cached) and Inkling Small at $0.50/$1.20 per 1M tokens ($0.10 cached).
Why it matters
Reserved-GPU price increases from a major neocloud are a leading indicator for compute costs across the AI stack. Watch whether other GPU cloud providers such as CoreWeave, Lambda, and Crusoe follow with their own reserved-tier increases in the coming weeks, and whether Togethers on-demand rate holds or eventually follows the reserved tiers upward.
In context
That the increase hits every reserved tier but leaves on-demand pricing untouched suggests reserved capacity, not spot demand, is where H100 supply is actually tight; providers typically raise on-demand rates first when demand is the binding constraint. Layering in two new serverless models the same period is a bundling move to keep customers engaged with the platform even as compute becomes more expensive to reserve.