Updated monthly

Changelog

New models, features, and API changes shipped to Venice each month.

Changelog

May 2025

Models

11 changes

Dolphin Mistral 24B Venice Edition
A new uncensored model added to the Venice lineup, now available as part of the curated model selection.
All users
Llama 4 Maverick
A new vision-enabled model with a 256K token context window, now available in the Venice model lineup.
All users
Venice Large
Venice Large is now powered by Qwen3 235B, replacing Llama4 Maverick. Beta users have shown strong preference for this new model.
All users
Venice Small
Venice Small is now powered by Qwen3 4B, replacing Llama 3.2 3B. Beta users have responded very positively to this upgrade.
All users

HiDream
HiDream achieves industry-leading scores on GenEval and DPG benchmarks, outperforming all other open-source models. It supports effective text placement and is commercial-friendly for personal and research use.
All users

47 changes

Avatar Popover
Introduced a character avatar popover on larger screens, displaying detailed character information and a quick chat option.
Character Creation Info
Added display of character creation date and creator information on public character profiles.
Message Timestamps
A setting has been added to enable display of message timestamps in App Settings.
Reasoning Toggle
A Reasoning toggle has been added for reasoning models that support enabling or disabling thinking responses.
Simple Mode
Simple Mode provides a streamlined experience by automatically routing prompts without requiring manual selection of chat type or model. It includes image context for building images through subsequent prompts.
Pro users
Temperature Warning
Added guidance when the Temperature setting is set very high, indicating potential gibberish output.

Enhance Only Mode
Support for 'Enhance Only' mode has been added, allowing enhance without changing the image resolution.
Hugging Face Model Card Link
Added a link to the Hugging Face model card from within the Image Detail view for better transparency and access to model details.
Image Loading State
An image generation loading state has been added to make it clearer that the UI is waiting on inference.
SD35 Enhance Guidance
Added in-app guidance to use descriptive prompts or the enhance feature when using the Venice SD35 image model.
Upscale & Enhance
Pro users can now upscale and enhance any image. Upscale maintains style and composition while increasing resolution. Enhance adds creative details and refinements beyond the original.
Pro users
Image Prompt Enhancer
Adjusted the image prompt enhancer to ensure its output stays within the character limit for image generation.

Auto-generate Confirmation
Added a confirmation box when using the auto-generate character feature to prevent accidental overwriting of existing character details.
Character Name Limit
Added a limit to character names to prevent UI issues caused by excessively long names.
Source Character Cloning
Added a UI feature to show which source character a user's character was cloned from.
Character UI
Updated the Character UI to display import and rating stats directly on primary character cards.

Crypto Renewal Flow
Added a subscription renewal flow for users paying with crypto who wish to renew.
Token Dashboard Coloring
Updated the token dashboard coloring for improved visual clarity.

Crypto Payment Support
Added support for purchasing API credits using cryptocurrency via Coinbase Commerce.
Local Storage Prompt
A prompt is now shown to permit the browser to persist local storage when storage is low.
Response Format Support
Enabled support for response_format in 405B and Deepseek models. Ensures thinking models don't inject <thinking> tags into the response when response format is enabled.
Space Bar Accept
Pressing the space bar now triggers the 'accept' button in confirmation screens for improved keyboard navigation.
Venice Search V2
Launched Venice Search V2, a complete overhaul of the search function that uses AI to generate context-aware queries, injects more results with richer data, and only shows sources actually referenced in responses via superscript citations.
Inference Error Messages
Improved error messages when there are issues communicating with Venice’s inference servers.

Mobile Wallet Redirect
Identifiable mobile wallets are now redirected to the token dashboard when accessing https://venice.ai, and the PWA installation modal is hidden.

25 changes

disable_thinking
Added support for disable_thinking, which enables strip_thinking_response automatically on reasoning models. API and model feature docs have been updated.
enable_web_citations
Added support for enable_web_citations to instruct the LLM to reference citations used in responses when Web Search is enabled. API and model docs updated.
gpt-4.1 Mapper
Added a model compatibility mapper for gpt-4.1 to map to Venice Large / Qwen 3 235B for improved interoperability.
Logprobs Support
Added support for the logprobs parameter on the /chat/completions API. API docs have been updated.
strip_thinking_response
Added support for strip_thinking_response parameter on reasoning models to suppress thinking blocks (<>), preventing them from reaching the client. Works with Qwen3 models.
Tools Function Calling
Added support for Tools and Function Calling to the Maverick model.
Beta

12 changes

Code Block Editing
Fixed a bug where editing messages with code blocks caused improper escaping of certain characters.
Chat
Context Filtering Logic
Refactored context filtering logic to prevent 'Your message is too long for the selected model' errors when dealing with chat conversations with large histories.
Chat
Document/Image Upload
Fixed a bug that prevented non-Pro users from uploading documents or images for analysis.
Chat
Edit Message Crash
Fixed a crash that occurred when editing a chat message and using the Space key.
Chat
Input History Navigation
Fixed a bug where the Input History Navigation setting did not properly control feature behavior.
Chat
Scrolling Bug
Fixed a scrolling bug affecting users with character chats.
Chat

Share feature requests and vote on ideas from the community on Learn Venice.