Skip to content
Back
Pricing briefing
TTogether
Together
PACKAGING CHANGE

Together rebuilds fine-tuning pricing from 3 size tiers to 29 per-model rows

By , Co-founder, PricingSaaS

Published

Together’s pricing page · Change detected

What changed

Together restructured its Fine-Tuning pricing page, replacing three size-tier buckets (Up to 16B, 17B-69B, 70-100B parameters) with a table of 29 individually named and priced models. The pricing tabs were relabeled from Standard/Specialized to LoRA/Full fine-tuning, and the flat $4.00 minimum job charge was replaced with per-model minimums ranging from $4.00 to $60.00.

Why it matters

For buyers, fine-tuning costs on Together now depend on which specific model is chosen rather than which size bucket it falls into, so cost comparisons that used to work off three tiers need to be redone per model, and job minimums as high as $60 change the math for small fine-tuning runs. For the category, it signals that as model catalogs balloon past two dozen options, size-tier pricing stops scaling and providers default to per-model tables instead — watch whether Together's ~8-day change cadence continues into Q4, and whether rival model marketplaces follow with similarly granular fine-tuning pricing.

In context

This is Together's fourth pricing restructure since reshuffling its Serverless Inference lineup in March, following July's Provisioned Throughput pricing overhaul, and it lands just one day after a 60% Qwen3.7-Max price hike — part of a run of 18 changes in 90 days. Restructures aren't rare across the index: 749 of the last 5,456 tracked changes (13.7%) involved one, with Datadog and Twilio the most frequent restructurers. The push toward per-model granularity echoes this week's repricing at DeepSeek, which just reversed a model retirement to keep billing steady past its own deadline. For Together, replacing broad size buckets with 29 individually priced models suggests its fine-tuning catalog has outgrown tier-based pricing.