Fireworks Adds GLM 5.3 to Serverless Training — Its Fourth Catalog Update Since Launch
Fireworks added the GLM 5.3 model (262K context) to its Serverless Training API pricing table.
Fireworks’s pricing page · Change Detected
WHY IT MATTERS
Training-infrastructure vendors are racing to keep per-token catalogs current with each new open-model release, and Fireworks is doing it while simultaneously raising the cost of raw compute — On-Demand GPU pricing rose 11-30% earlier this month. The two moves together point to a company monetizing convenience (managed per-token training) more aggressively than capacity. Watch whether Fireworks keeps this weekly cadence on the Training API or settles back toward its historical monthly rhythm.
BY THE NUMBERS
- 9 days since Fireworks' last pricing change (Sep 7)
- 8 changes in the last 90 days vs. ~30-day average cadence
- Fourth Serverless Training API catalog update since its Aug 3 launch
- 605 credits/usage-billing changes across 336 companies in the last 180 days
THEIR RATIONALE
This is Fireworks' fourth change to the Serverless Training API catalog since the product launched on Aug 3, and it comes just 9 days after the company last rebuilt the same catalog (dropping Qwen 3.5 9B, adding DeepSeek V4 Flash and Muse Glimmer 30B). The framing is straightforward model-lineup expansion — keeping the per-token training menu current with newly released open models rather than a pricing overhaul.
THE SIGNAL
The single-model addition understates how fast this product line is moving: Fireworks has logged 8 pricing changes in the last 90 days against its own ~30-day average cadence, and nearly half of the recent changes touch this one API. That's part of a wider shift toward metered, usage-based pricing across the index — 605 changes from 336 companies in the last 180 days carry a credits-or-usage-billing pattern, and this is Fireworks' tenth entry in that pattern since tracking began.
IN CONTEXT
Fireworks has touched its Serverless Training API catalog four times since launching it on Aug 3, most recently just nine days after rebuilding the same catalog to drop Qwen 3.5 9B and add DeepSeek V4 Flash and Muse Glimmer 30B. That's well above the company's own baseline of roughly one change every 30 days, and it lands alongside an 11-30% On-Demand GPU price hike from earlier this month — Fireworks is expanding its token-priced training surface even as raw compute gets more expensive. It's a smaller, faster-cycling version of a pattern playing out across the index: 605 credits-and-usage-billing changes from 336 companies in the last 180 days, led by movers like HubSpot and Datadog.