Fireworks expands Serverless Training with Muse Spark Open and DeepSeek V4.1 Flash
Fireworks' Serverless Training API table gained two new base models.
Fireworks’s pricing page · Change Detected
WHY IT MATTERS
Teams fine-tuning on Fireworks now have DeepSeek V4.1 Flash as a materially cheaper training option alongside pricier models like Muse Spark Open, within the same transparent table that contrasts with Fireworks' now-hidden serverless inference prices. Watch whether Fireworks keeps Serverless Training fully priced in public as it keeps adding models, or eventually treats it the way it treated inference pricing in May.
BY THE NUMBERS
- Latest Serverless Training addition, 19 days after GLM 5.3 Flash
- 7 real pricing changes in the last 90 days, vs. a 34-day average gap
- DeepSeek V4.1 Flash costs $4.98 per 1M to train, vs. $18.58 for Muse Spark Open
- 21 feature-added events, Fireworks' single most common change type
THEIR RATIONALE
Fireworks is steadily filling out its pay-per-token LoRA training product rather than repricing what's already there: this is another addition to Serverless Training, following GLM 5.3 Flash 19 days ago. Fireworks has made 7 real pricing changes in the last 90 days, against a typical gap of about 34 days between changes, so the pace here is faster than its historical average.
THE SIGNAL
Feature and model additions dominate Fireworks' track record — 21 features added events across its history, its single most common change type — and Serverless Training itself is a newer pay-per-token LoRA product Fireworks keeps building out. That contrasts with Fireworks' separate retreat from transparent serverless inference pricing in May 2026, when per-model token prices were pulled from the public page entirely. Training pricing, by comparison, stays fully listed and is where Fireworks keeps adding rows. Market-wide, 201 pricing changes across 153 companies have touched credits-and-usage billing in the last 180 days, 11.6% of everything tracked, so a growing per-token training catalog fits the wider shift toward metered AI infrastructure pricing.
IN CONTEXT
This is Fireworks' latest addition to Serverless Training, coming 19 days after it added GLM 5.3 Flash with 200K context to the same per-token training table. Fireworks has made 7 real pricing changes in the last 90 days, faster than its roughly 34-day average gap between moves. The contrast with May's decision to pull serverless inference prices from the public page entirely is notable: Training pricing stays fully listed even as the catalog grows. Market-wide, 201 pricing changes across 153 companies have touched credits-and-usage billing in the last 180 days, 11.6% of everything tracked.