Together AI pricing page — GPU Clusters on-demand, Preemptible Compute and reserved tier table (re-read 2026-10-05)

Observed 2026-10-05. The page carries THREE different GPU tables with DIFFERENT price levels for the same silicon, and the repo's together-gpu-clusters-h100 record cites only one of them. This source exists to make the tier structure explicit. (1) 'Dedicated Inference' table, headed 'All prices per gpu per hour': HGX H100 $5.49, HGX B200 $8.99, HGX H200 / B300 / GB200 NVL72 / GB300 NVL72 all 'Contact us'. THIS is the figure the existing record stores as price_usd=5.49, and it is correct for that table. (2) 'GPU Clusters — On-demand' table ('Pay as you go GPU capacity on an hourly basis'): HGX H100 $3.99, HGX H200 $5.99, HGX B200 $8.19, HGX B300 $9.99, GB200 NVL72 and GB300 NVL72 '—'. (3) 'On-demand hourly rates and reserved capacity' table ('All prices are per GPU per hour'), columns Preemptible Compute Pay-as-you-go | ON-Demand Pay-as-you-go | Reserved 7-30 days | 31-90 days | 91-180 days | 181+ days: HGX H100 $1.99 | $3.99 | $3.69 | $3.45 | $3.19 | Contact us; HGX H200 $2.99 | $5.99 | $4.99 | $4.15 | $3.99 | Contact us; HGX B200 $4.09 | $8.19 | $7.99 | $7.79 | $6.79 | Contact us; HGX B300 $4.99 | $9.99 | Contact us for all reserved terms; GB200 NVL72 and GB300 NVL72 '—'. THE PRACTICAL CONSEQUENCE: at Together, an H100 GPU-hour costs $5.49 on the Dedicated Inference product and $3.99 on the GPU Clusters on-demand product — a 1.38x spread on identical silicon and an identical per-GPU-hour unit, differing only in which product you are buying. That is the tier trap the repo's own cost_per_token tier column exists to catch, and the existing record does not carry the Clusters figures at all. Note also that $5.49 in the record is the DEAREST H100 number on the page, and the cheapest published H100 number on the same page is the $1.99 preemptible rate.