Together hikes Qwen3.7-Max prices up to 92%, drops LFM2.5-8B again
By Rob Litterst, Co-founder, PricingSaaS
Published
Together’s pricing page · Change detected
What changed
Together raised Qwen3.7-Max serverless inference pricing across the board: input $1.25 -> $2.00 per 1M tokens (+60%), cached input $0.13 -> $0.25 (+92%), and output $3.75 -> $6.00 (+60%). In the same catalog update, the LFM2.5-8B-A1B model ($0.03 input / $0.12 output per 1M tokens) was removed from the Serverless Inference chat model catalog.
Why it matters
For teams routing workloads through Together's serverless inference, the Qwen3.7-Max jump is a reminder that flagship-model pricing on aggregator platforms can move sharply with no advance notice - a 60-92% swing changes the unit economics of anything built on that model. Watch whether Together's next catalog update, likely within its roughly 8-day average cadence, brings another increase on a different flagship model or whether this settles as a one-off Qwen3.7-Max repricing.
In context
Together's price sheet rarely sits still: this is its 31st tracked change since Together first appeared in the index on Dec 29, 2025, and its second price increase in six weeks after the Aug 3 GPU rate hike, arriving just ten days after Qwen3.8 Flash was added on Sep 1. The 60-92% jump on Qwen3.7-Max is steeper than Together's other 2026 increases, including the 37-40% GLM-5.1 hike in April. Price increases are still a minority move industry-wide — 211 of 5,452 tracked changes in the last 180 days, versus far more common usage-billing shifts like Replit's Sep 11 Starter plan cut or Mintlify's credit-cost reveal. LFM2.5-8B-A1B's exit, meanwhile, continues a pattern of the model cycling in and out of the catalog since it was added Jul 20.