Reasoning guide added; cached prompt tokens on serverless cost 50% less

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

📚 Documentation Updates

• Reasoning Guide: New documentation for working with reasoning models, including reasoning_content, reasoning_effort, streaming, and interleaved thinking (Reasoning)
• Recommended Models: Updated recommendations to include DeepSeek V3.2 for code generation and Kimi K2 Thinking as a GPT-5 alternative (Recommended Models)
• OpenAI Compatibility: Removed stop sequence documentation as Fireworks is now 1:1 compatible with OpenAI's behavior (OpenAI Compatibility)
• Evaluator APIs: Added REST API documentation for Evaluator and Evaluation Job CRUD operations
• firectl CLI Reference: Updated with new commands including cancel dpo-job, cancel supervised-fine-tuning-job, set-api-key, redeem-credit-code, and evaluator revision management

[accordion] Bug Fixes & Minor Improvements

• Audio Usage Charts: Added a Voice modality tab to the Usage page for viewing audio-specific usage metrics and charts (Web App)
• Deployment Page Redesign: Redesigned deployment details page with inline action buttons (Edit, Delete, Enable/Disable), reordered metadata sections, and collapsible API examples (Web App)
• Deployment API Examples: API code examples now use the deployment route as the canonical model identifier for clearer usage patterns (Web App)
• LoRA Addons Tab: Renamed "Serverless LoRA" tab to "LoRA Addons" on the deployments dashboard for clarity (Web App)
• Evaluator Selector: Improved evaluator selector UI to display both display name and evaluator ID for easier identification (Web App)
• Evaluator Delete Confirmation: Added confirmation modal with success/error feedback when deleting evaluators (Web App)
• Evaluator Code Viewer: Evaluator source files now load asynchronously, preventing browser freezes with large files (Web App)
• Dataset Image Preview: Dataset preview now properly renders image content in message bubbles and comparison views (Web App)
• Model Search: Improved model search accuracy by restricting results to display name and model ID matches only (Web App)
• Fine-Tuning Progress Status: Fine-tuning job detail pages now show the initial job status immediately on page load (Web App)
• Evaluator Test Controls: Dataset selection and pagination are now disabled while an evaluator test is running to prevent conflicts (Web App)
• Repository Name Validation: Evaluator repository names now validate against GitHub naming conventions (Web App)
• Billing Contracts Tab: Added a Contracts tab to the billing page alongside Invoices for viewing contract details (Web App)
• Dataset Size Limit: Enforced 1GB maximum file size for dataset uploads with clear error messaging (Web App)
• Evaluator Documentation Link: Fixed evaluator documentation links to point to the correct location (Web App)
• Session Update Fix: Fixed an issue with session state updates in the web app (Web App)
• Popover Fix: Fixed popover components nested in dialogs not displaying correctly (Web App)
• Per-Replica Status: Deployment status now shows per-replica counts (pending, downloading, initializing, ready) for better visibility into deployment progress (CLI, API)
• set-api-key Command: The set-api-key command is now visible in firectl help output (CLI)
• DPO Resume: Added support for resuming cancelled DPO fine-tuning jobs (API)
• Reasoning Effort Normalization: reasoning_effort parameter now accepts boolean values in addition to strings and integers (API)