Fireworks AI Release Notes

<Badge color="blue">Inference</Badge>

Serverless deprecation: Kimi K2.5 and Qwen 3.6 Plus

Kimi K2.5 and Qwen 3.6 Plus are deprecated from serverless.

Recommended migrations

• Kimi K2.5 — migrate to Kimi K2.6
• Qwen 3.6 Plus — migrate to Qwen 3.7 Plus

Several legacy serverless models will be decommissioned on May 14, 2026 to make room for newer, higher-performance releases. This applies only to serverless usage; dedicated deployments are unaffected.

Action required

If you use any of the models below on serverless, migrate to a recommended replacement before May 14, 2026. After that date, they will no longer be available via serverless endpoints.

Serverless migration guide

• DeepSeek V3.1 — migrate to Kimi K2.6 or GLM 5.1
• DeepSeek V3.2 — migrate to Kimi K2.6 or GLM 5.1
• GLM 4.7 — migrate to GLM 5.1
• GLM 5 — migrate to GLM 5.1
• Qwen3 8B — migrate to GPT-OSS 20B
• Qwen3 VL 30B A3B Thinking — migrate to Kimi K2.6
• Qwen3 VL 30B A3B Instruct — migrate to Kimi K2.6
• Llama 3.3 70B Instruct — migrate to GPT-OSS 120B …

<Badge color="blue">Inference</Badge>

Serverless deprecation: legacy models removed May 14, 2026

Several legacy serverless models will be decommissioned on May 14, 2026 to make room for newer, higher-performance releases. This applies only to serverless usage; dedicated deployments are unaffected.

Action required

If you use any of the models below on serverless, migrate to a recommended replacement before May 14, 2026. After that date, they will no longer be available via serverless endpoints.

Serverless migration guide

• DeepSeek V3.1 — migrate to Kimi K2.6 or GLM 5.1
• DeepSeek V3.2 — migrate to Kimi K2.6 or GLM 5.1
• GLM 4.7 — migrate to GLM 5.1
• GLM 5 — migrate to GLM 5.1
• Qwen3 8B — migrate to GPT-OSS 20B
• Qwen3 VL 30B A3B Thinking — migrate to Kimi K2.6
• Qwen3 VL 30B A3B Instruct — migrate to Kimi K2.6 …

<Badge color="blue">Inference</Badge> <Badge color="purple">Training</Badge> <Badge color="gray">Platform</Badge>

Video & Audio Models, AWS S3 Training Integration, and SSO Improvements

Video & Audio Input Models

You can now query multimodal models with video and audio inputs for video captioning, scene analysis, and multimodal question answering. Deploy models like Qwen3 Omni and Molmo2 to process video and audio content directly using the Chat Completions API.

See the Video & Audio Inputs guide for deployment instructions and code examples.

AWS S3 Integration for Training Datasets

Training datasets can now be stored in your own AWS S3 buckets using GCP-to-AWS OIDC federation. This Bring Your Own Bucket (BYOB) approach keeps your data private while enabling secure access during Supervised Fine-Tuning and Reinforcement Fine-Tuning jobs—no long-lived credentials required.

See the Secure Training (BYOB) documentation for IAM role setup and usage examples.

Just-In-Time (JIT) User Provisioning for SSO (Enterprise)

JIT user provisioning automatically creates user accounts when users sign in through SSO for the first time. Enable this when configuring your identity provider to eliminate manual user creation.

See the SSO documentation for setup instructions.

📚 Documentation Updates …

<Badge color="purple">Training</Badge> <Badge color="gray">Platform</Badge>

Warm-Start Training and Azure Model Uploads

Warm-Start Training for Reinforcement Fine-Tuning

You can now warm-start Reinforcement Fine-Tuning jobs from previously supervised fine-tuned checkpoints using the --warm-start-from flag. This enables a streamlined SFT-to-RFT workflow where you first train a model with supervised fine-tuning, then continue training with reinforcement learning.

See the Warm-Start Training guide for details.

Azure Federated Identity for Model Uploads

Model uploads from Azure Blob Storage now support Azure AD federated identity authentication as an alternative to SAS tokens. This eliminates the need for credential rotation and enables secure, credential-less authentication.

See the Uploading Custom Models documentation for setup instructions.

📚 Documentation Updates

• Warm-Start Training: New guide for SFT-to-RFT workflows (Warm-Start Training)
• Azure Federated Identity: Setup instructions for Azure AD authentication (Uploading Custom Models)
• Preserved Thinking: Multi-turn reasoning with preserved thinking context (Reasoning)
• GLM 4.7: Added to models supporting reasoning_effort parameter …

<Badge color="blue">Inference</Badge> <Badge color="purple">Training</Badge> <Badge color="gray">Platform</Badge>

Playground Categories, New User Roles, Fine-Tuning Improvements, and New Models

Playground Categories

The Playground now features category tabs (LLM, Image, TTS, STT) in the header for easier switching between model types. The playground automatically detects the appropriate category based on the selected model and provides smart defaults for each category.

User Roles: Contributor and Inference

New user roles provide more granular access control for team collaboration:

• Contributor: Read and write access to resources without administrative privileges
• Inference: Read-only access with the ability to run inference on deployments

Assign these roles when inviting team members to provide appropriate access levels.

Fine-Tuning Improvements

Fine-tuning workflows have been enhanced with several new capabilities:

• Stop and Resume Jobs: Stop running fine-tuning jobs and resume them later from where they left off. Available for Supervised Fine-Tuning and Reinforcement Fine-Tuning jobs.
• Clone Jobs: Quickly create new fine-tuning jobs based on existing job configurations using the Clone action.
• Download Output Datasets: Download output datasets from Reinforcement Fine-Tuning jobs, including individual files or bulk download as a ZIP archive. …

<Badge color="blue">Inference</Badge> <Badge color="purple">Training</Badge> <Badge color="gray">Platform</Badge>

Reasoning Guide, Prompt Caching Updates, New Models and CLI Updates

Reasoning Guide

A new Reasoning guide is now available in the documentation. This comprehensive guide covers:

• Accessing reasoning_content from thinking/reasoning models
• Controlling reasoning effort with the reasoning_effort parameter
• Streaming with reasoning content
• Interleaved thinking for multi-step tool-calling workflows

The guide provides code examples using the Fireworks Python SDK and explains how to work with models that support extended reasoning capabilities.

Prompt Caching Updates

Prompt caching documentation has been updated with expanded guidance:

• Cached prompt tokens on serverless now cost 50% less than uncached tokens
• Session affinity routing via the user field or x-session-affinity header for improved cache hit rates
• Prompt optimization techniques for maximizing cache efficiency

See the Prompt Caching guide for details.

✨ New Models

• Devstral Small 2 24B Instruct 2512 is now available in the Model Library
• NVIDIA Nemotron Nano 3 30B A3B is now available in the Model Library …

<Badge color="blue">Inference</Badge> <Badge color="purple">Training</Badge> <Badge color="gray">Platform</Badge>

DeepSeek V3.2 on Serverless, Cached Token Pricing, and New Models

☁️ Serverless

• DeepSeek V3.2 is now available on serverless

Cached Token Pricing Display

The Model Library and model detail pages now display cached and uncached input token pricing for serverless models that support prompt caching. This gives you better visibility into potential cost savings when using prompt caching with supported models.

Evaluations Dashboard Improvements

The Evaluations dashboard has been enhanced with new filtering and status tracking capabilities:

• Status column showing evaluator build state (Active, Building, Failed)
• Quick filters to filter evaluators and evaluation jobs by status
• Improved table layout with actions integrated into the status column

✨ New Models

• DeepSeek V3.2 is now available in the Model Library
• Ministral 3 14B Instruct 2512 is now available in the Model Library
• Ministral 3 8B Instruct 2512 is now available in the Model Library …

<Badge color="blue">Inference</Badge> <Badge color="purple">Training</Badge> <Badge color="gray">Platform</Badge>

Audit Logs, Dataset Download, Weighted Training for Reinforcement Fine-Tuning, and New Model

Audit Logs in Web App

You can now view and search audit logs directly from the Fireworks web app. The new Audit Logs page provides:

• Search and filter logs by status and timeframe
• Detailed view panel for individual log entries
• Easy navigation from the console sidebar under Account settings

See the Audit Logs documentation for more information.

Dataset Download

You can now download datasets directly from the Fireworks web app. The new download functionality allows you to:

• Download individual files from a dataset
• Download all files at once with "Download All"
• Access downloads from the Datasets table in the dashboard

Weighted Training for Reinforcement Fine-Tuning

Reinforcement Fine-Tuning now supports per-example weighting, giving you more control over which samples have greater influence during training. This feature mirrors the weighted training functionality already available in Supervised Fine-Tuning.

See the Weighted Training documentation for details on the weight field format.

✨ New Models …

<Badge color="blue">Inference</Badge> <Badge color="purple">Training</Badge> <Badge color="gray">Platform</Badge>

Evaluator Improvements, Kimi K2 Thinking on Serverless, and New API Endpoints

Improved Evaluator Creation Experience

The evaluator creation workflow has been significantly enhanced with GitHub template integration. You can now:

• Fork evaluator templates directly from GitHub repositories
• Browse and preview templates before using them
• Create evaluators with a streamlined save dialog
• View evaluators in a new sortable and paginated table

MLOps & Observability Integrations

New documentation for integrating Fireworks with MLOps and observability tools:

• Weights & Biases (W&B) integration for experiment tracking during fine-tuning
• MLflow integration for model management and experiment logging

✨ New Models

• Kimi K2 Thinking is now available in the Model Library
• KAT Dev 32B is now available in the Model Library
• KAT Dev 72B Exp is now available in the Model Library

☁️ Serverless

• Kimi K2 Thinking is now available on serverless

📚 New REST API Endpoints …

<Badge color="purple">Training</Badge>

☀️ Sunsetting Build SDK

The Build SDK is being deprecated in favor of a new Python SDK generated directly from our REST API. The new SDK is more up-to-date, flexible, and continuously synchronized with our REST API. Please note that the last version of the Build SDK will be 0.19.20, and the new SDK will start at 1.0.0. Python package managers will not automatically update to the new SDK, so you will need to manually update your dependencies and refactor your code.

Existing codebases using the Build SDK will continue to function as before and will not be affected unless you choose to upgrade to the new SDK version.

The new SDK replaces the Build SDK's LLM and Dataset classes with REST API-aligned methods. If you upgrade to version 1.0.0 or later, you will need to migrate your code.

🚀 Improved RFT Experience

We've drastically improved the RFT experience with better reliability, developer-friendly SDK for hooking up your existing agents, support for multi-turn training, better observability in our Web App, and better overall developer experience.

See Reinforcement Fine-Tuning for more details.

<Badge color="purple">Training</Badge>

Supervised Fine-Tuning

We now support supervised fine tuning with separate thinking traces for reasoning models (e.g. DeepSeek R1, GPT OSS, Qwen3 Thinking etc) that ensures training-inference consistency. An example including thinking traces would look like:

  {
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "What is the capital of France?"}, 
      {"role": "assistant", "content": "Paris.", "reasoning_content": "The user is asking about the capital city of France, it should be Paris."}
    ]
  }
  {
    "messages": [
      {"role": "user", "content": "What is 1+1?"},
      {"role": "assistant", "content": "2", "weight": 0, "reasoning_content": "The user is asking about the result of 1+1, the answer is 2."},
      {"role": "user", "content": "Now what is 2+2?"},
      {"role": "assistant", "content": "4", "reasoning_content": "The user is asking about the result of 2+2, the answer should be 4."}
    ]
  }

We are also properly supporting multi-turn fine tuning (with or without thinking traces) for GPT OSS model family that ensures training-inference consistency.

<Badge color="purple">Training</Badge> <Badge color="gray">Platform</Badge>

🎨 Vision-Language Model Fine-Tuning

You can now fine-tune Vision-Language Models (VLMs) on Fireworks AI using the Qwen 2.5 VL model family. This extends our Supervised Fine-tuning V2 platform to support multimodal training with both images and text data.

Supported models:

• Qwen 2.5 VL 3B Instruct
• Qwen 2.5 VL 7B Instruct
• Qwen 2.5 VL 32B Instruct
• Qwen 2.5 VL 72B Instruct

Features:

• Fine-tune on datasets containing both images and text in JSONL format with base64-encoded images
• Support for up to 64K context length during training
• Built on the same Supervised Fine-tuning V2 infrastructure as text models

See the VLM fine-tuning documentation for setup instructions and dataset formatting requirements.

🔧 Build SDK: Deployment Configuration Application Requirement

The Build SDK now requires you to call .apply() to apply any deployment configurations to Fireworks when using deployment_type="on-demand" or deployment_type="on-demand-lora". This change ensures explicit control over when deployments are created and helps prevent accidental deployment creation.

Key changes:

• .apply() is now required for on-demand and on-demand-lora deployments
• Serverless deployments do not require .apply() calls …