# Fireworks AI makes its Training API generally available for custom model training

> Original title: Train past the frontier: Training API now generally available

- Source: Fireworks AI Blog
- Published: 2026-08-31T00:00:00.000Z
- Bad Influence: https://badinfluence.news/items/lbwemczvrp5a1xiws60oxkmui
- Original: https://fireworks.ai/blog/training-api-general-availability/

## Summary

Fireworks AI announces general availability of its Training API and Fireworks Lab. The Training API lets teams run their own Python training loop on Fireworks-managed trainer and rollout infrastructure, with control over loss, reward, data, and environment. Fireworks says it aligns numeric formats and kernels between training and inference and hot-loads updated weights, with checkpoint transfers reduced by up to 10x.

## Why it matters

The post explains how the Training API connects custom loops to managed distributed training, and how numeric alignment and weight hot-loading address common RL bottlenecks.