# Dify 0.9.0 — Release & Upgrade Intelligence
- **Product:** Dify (https://dify.ai)
- **Version:** 0.9.0
- **Released:** Sep 30, 2024
- **Impact Score:** 5/10
- **Breaking Changes:** Not verified
- **Canonical Analysis:** https://changelogs.us/dify/updates/0.9.0/

## Executive Summary (TL;DR)
Dify 0.9.0. Source highlight: ✨ What’s New in v0.9.0? ✨

## New Features
- **Voyage AI**: A new model provider added (thanks to @hwzhuhao in #8747).
- Introduced **InternLM 2.5-20b** and **Qwen 2.5-coder-7b** (@hwzhuhao in #8862).
- New **Nomic embedding model provider** introduced (@yaoice in #8640).
- **DeepSeek-v2.5** added as a model provider for SiliconFlow (@hwzhuhao in #8639).
- Added **Min-connections and Max-connections** settings for pgvector (@hwzhuhao in #8841).
- Added English versions for customizable model scale documentation to broaden accessibility (@hwzhuhao in #8871).

## Improvements
- **Llama3.2**: Now available with Fireworks and Groq providers for improved performance and variety (@yaoice in #8809 and @Howe829 in #8831).
- Improved **SVG output UX** for a more seamless design experience (@xuzuodong in #8765).
- Extended **Jina Embeddings** V3 API parameters (@DresAaron in #8657).
- Deprecated older **Mistral models** for SiliconFlow to streamline performance (@hwzhuhao in #8828).
- **Error Handling Enhancements**: Improved iteration filters and error logging for optimal performance (@hwzhuhao in #8461).
- Enhanced operational flexibility with fully configurable **WORKFLOW_* environment variables** (@fujita-h in #8644).

## Bug Fixes
- Fixed "Max submit count reached" error during workflow iterations (@takatost in #8595).
- Corrected a linking error in help documents (@hjlarry in #8508).
- **iOS TTS Stream**: Now fully functional on Safari 17.1+ (@likenamehaojie in #8645).
- Resolved JSONDecodeError in **llm_generator.py** (@ZuzooVn in #8504).
- Fixed errors when sending messages within chatflow and dialogs (@xuzuodong in #8627 and #8682).
- Corrected **Spark's LLM token calculation** error for accurate token usage (@cherryhuahua in #8755).