'%3e%3crect%20x='16'%20y='51'%20width='68'%20height='27'%20rx='13.5'/%3e%3ccircle%20cx='32'%20cy='53'%20r='15'/%3e%3ccircle%20cx='53'%20cy='43'%20r='20'/%3e%3ccircle%20cx='71'%20cy='57'%20r='12'/%3e%3c/g%3e%3c/svg%3e){kind=small}
Cloud Changelog
Fireworks AI platform changelog 2025
24 announcements, 2025
All Fireworks AI platform changes · 65 announcements since May 2025
Free. Unsubscribe any time.
RSS feed for Fireworks AI platform
December 20254
Playground category tabs, Contributor and Inference roles, and stop and resume for fine-tuning jobs
Reasoning guide added; cached prompt tokens on serverless cost 50% less
DeepSeek V3.2 on serverless, plus cached and uncached token pricing in Model Library
Audit Logs and dataset download in web app, weighted training for Reinforcement Fine-Tuning
November 20253
August 20252
July 20257
Fine-tuning for Qwen 2.5 VL vision-language models, with up to 64K context
Build SDK requires .apply() for on-demand and on-demand-lora deployments
Custom rollout and reward functions via LLM.reinforcement_step() and ReinforcementStep
Supervised fine-tuning supports Llama 4 Scout and Llama 4 Maverick, text only
Build SDK LLM class requires id for on-demand and base_id for on-demand-lora
Log in to Fireworks with LinkedIn, in the web app and with firectl login
June 20256
Log in to Fireworks with GitHub, in the web UI and with firectl login
Build SDK integrates with reward-kit for Reinforcement Fine-Tuning evaluators
Response API adds previous_response_id for multi-turn context, and streaming
Reinforcement Fine-Tuning released, training models through verifiable reward
Supervised Fine-Tuning V2 released, with multi-turn function calling tuning