'%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
65 announcements since May 2025
Free. Unsubscribe any time.
RSS feed for Fireworks AI platform
November 20263
October 202612
Rune 26B-A4B available on Serverless, returning probabilities for up to 256 questions
fireworks-ai 1.2.30 replaces TITO sidecar 502 errors with source-specific error codes
Six Serverless models deprecated, to be decommissioned on November 7, 2026
NVIDIA Nemotron 3.5 Super VL 120B A12B available for training, up to 262K context
DeepSeek V4 Flash (0731) to be removed from Serverless Training on November 2, 2026
Small serverless tier now covers models under 800B parameters, up from 600B
DeepSeek V4.1 Flash 1017 available for LoRA on Serverless Training, 262K context
Muse Spark Open available for LoRA on Serverless Training, up to 262K context
Small serverless model rate-limit ceilings doubled, to 216M total prompt TPM
Small serverless tier rate limits raised to 108M total prompt TPM, tier expanded
DeepSeek V4.1 Flash serverless pricing changes: output rises to $1.20 per 1M tokens
Managed reinforcement fine-tuning paused; 70 models no longer tunable
September 202612
DeepSeek V4.1 Flash available for LoRA training with up to 262K context
Serverless deprecation takes effect for DeepSeek V4, Kimi K2 and GLM 5.2 models
GLM 5.3 Flash available for LoRA on Serverless Training, with 200K context
Create Deployment adds deploymentShape default and acceptShapelessRisk options
GLM 5.3 available for LoRA on Serverless Training, with up to 262K context
GLM 5.3 Flash available for LoRA on Dedicated Training API, with vision
Fireworks serverless decommissions DeepSeek V4, GLM 5.2 and Kimi K2 models on Sept 25
Fireworks training skill added for Claude Code, Cursor and Codex agents
Training cost estimator added to estimate job costs before running
Deployment tags added in firectl 1.8.3; REST annotation keys require custom/ prefix
GLM 5.3 available for LoRA training on Dedicated Training API, up to 204K context
Serverless adaptive rate limit ceilings now vary by model size tier
August 20265
DeepSeek V4 Flash 0731, Qwen 3.8 27B and Muse Glimmer 30B added for LoRA training
MiniMax M2.7, GPT OSS 20B and three more models deprecated from serverless
Qwen 3.5 9B and Qwen 3.6 27B deprecated from Serverless Training on August 26, 2026
DeepSeek V4 Flash deprecated from serverless, migrate to DeepSeek V4 Flash (0731)