Together AI Pricing - GPU Clusters on-demand/preemptible/reserved + Dedicated Inference

Fetched 2026-10-04. This record SUPERCEDES the existing sup-cloud-together-gpu-clusters-pricing (which was based on an earlier fetch). Now shows full pricing with preemptible and reserved tiers. DEDICATED INFERENCE (per GPU-hour, single-tenant): HGX H100 $5.49 on-demand; HGX B200 $8.99 on-demand; H200, B300, GB200 NVL72, GB300 NVL72 all 'Contact us'. GPU CLUSTERS (on-demand hourly, per GPU): H100 $3.99, H200 $5.99, B200 $8.19, B300 $9.99. GPU CLUSTERS PREEMPTIBLE: H100 $1.99, H200 $2.99, B200 $4.09, B300 $4.99. GPU CLUSTERS RESERVED (7-30 days / 31-90 days / 91-180 days): H100 $3.69/$3.45/$3.19; H200 $4.99/$4.15/$3.99; B200 $7.99/$7.79/$6.79; B300 contact-sales. 181+ days all contact-sales. PROVISIONED THROUGHPUT: $0.05/PTU/min, where PTU is model-specific throughput units (e.g. MiniMax M3: 166,667 input TPM/PTU, 41,667 output TPM/PTU). Serverless inference priced per 1M tokens (separate unit, not recorded as supply). Key observation: Together's on-demand H100 at $3.99/GPU-hr is significantly below the CoreWeave on-demand rate of $6.16/GPU-hr, and their preemptible at $1.99 is the cheapest published H100 rate we've seen from a managed provider.