Changelog

Venice Now Hosts All Leading Frontier Models

We've significantly expanded our model library to include the latest and most powerful frontier AI models available. For text, we've added Grok 4.1 Fast with reasoning and vision, Claude Opus 4.5, Gemini 3 Pro Preview, Kimi K2, GPT-5.2, and Deepseek 3.2. For video, we've launched LTX 2 Video, Kling 2.6, and the new Longcat video models. For image generation, GPT-Image-1.5 is now available for alpha users.

These models use credits, and requests through these models are anonymized. Venice is now your one-stop-shop for accessing cutting-edge AI.

App

• Fixed a bug where upscaling or enhancing an image at 1x would switch the conversation type to "image". The conversation type now remains consistent.

• The "Use original prompt" option for upscaling is now correctly capped at the maximum prompt length for the enhance feature.

• Added the ability to mute users directly in the Social Feed.

• The "Auto" model is now hidden from the model switcher when you are in the video category.

• Fixed an issue where credit estimates were not displaying correctly on small screens.

• Added resolution selection to the NanoBananaPro model.

• Added an overflow scroll to the memory documents section to prevent UI issues.

• Moderated posts now appear at the top of the Social Feed for visibility.

• Added a tooltip to the locked label on the token dashboard for better clarity.

• Fixed an overflow issue on the preferences modal for guest users on the profile page.

• Added feedback buttons to the Support Bot responses.

• Hid the mature filter in preferences for non-Pro users.

• Improved system-prompt token calculation for more accurate usage reporting.

• Fixed a bug in multimodal regen.

• Swapped user-facing "Beta" tags to "Alpha" or "Alpha Tester" for consistency.

• Added estimated pricing display for Pay Per Use LLM models in the model selector.

• Added drag-and-drop image upload functionality to the edit, combine, and upscale modals.

• Replaced all instances of "points" with "Venice Credits" in the user profile and settings.

• Relaxed schema validation for chat completions to allow unknown fields, improving compatibility with third-party clients.

• Updated the chat input loading spinner and image loading animation for a smoother experience.

• Added a way to feature characters and updated the default sort on the public characters page to "Featured".

• Fixed duplicate entries on the "Browse Characters" page.

• Fixed the NSFW overlay regeneration logic.

• Updated search provider labels for clarity.

• Added resolution and websearch information to the image details view.

• Fixed style preset settings for models that do not support them.

• Added a message to the Support bot for non-Venice related questions.

• Added a copy button for markdown tables.

• Fixed buttons in the row component from shrinking incorrectly.

Mobile App

• Added 3 new languages (using Google Translate for now).

• Mobile credit purchases now redirect back to the native app after completion.

• Fixed tooltips getting stuck on Android devices.

Models

• Promoted nano banana pro to the beta API.

• Increased the price of Nano Banana Pro.

• Launched the new LTX 2 Video model.

• Added the new Grok 4.1 Fast model to all users and the API, with reasoning and vision enabled.

• Added new Pay Per Use models: Gemini 3 Pro Preview and Kimi K2.

• Upgraded Seedream to version 4.5.

• Launched Kling 2.6 publicly.

• Launched Deepseek 3.2 as a live PRO model, with API support and reasoning enabled.

• Added new Longcat video models.

• Added the new Claude Opus 4.5 model.

• Made GPT-Image-1.5 live for alpha users.

• Removed DeepSeek R1 (beta) from the API.

• Began the two-step process of retiring the Venice-hosted Qwen235B model.

• Added the new GPT-5.2 model.

• Removed Devstral 2 from the API and Kandinsky 5 from the app.

• Enabled reasoning for Deepseek 3.2 again after provider fixes.

API

• Added Fal image generations to the API.

• Updated API documentation for FAL model configurations.

• Added support for the reasoning_effort parameter.

• Added support for the cache_control parameter.

• Added support for prompt_cache_key in the LLM schema to improve cache hits.

• Preserved reasoning_details blocks for advanced tool calling and reasoning.

• Fixed API Dashboard usage display to handle negative values correctly.

• Added support for analysing audio and video files for Gemini 3 Pro/Flash and other compatible models.

• Fixed API KEY consumption, which was incorrectly limiting some users to 50% of their available Diem balance.

• Added consumable balance to video API request headers.

• Added support for Anthropic-format tool configurations and the tool_use/tool_result message format.

• Added cache token tracking and discounted billing for models that support context caching, including Anthropic models.

Token

• Fixed the diem_global_utilization view to include refund entries.

• Busted the Diem usage chart cache to reflect corrected data.

• Added a new "burn" section to the token page.

Fixes and Improvements

• Fixed a bug where prompts longer than the enhance model's limit would cause errors.

• Fixed credit estimates display on small screens.

• Fixed incorrect display of the "Reasoning" tag in text details.

• Fixed an image aspect ratio issue in multimodal chats.

• Fixed resolution examples to use "1K", "2K", and "4K" labels.

• Fixed social user name validation to allow hyphens.

• Fixed FAL image model pricing.

• Fixed regen for character images.

• Fixed aspect ratio issues in selfie mode and the image loader.

• Fixed a bug where the info and feedback buttons were hidden in grid view.

New Pay-Per-Use image models: Nano Banana, Nano Banana Pro, Seedream v4

We launched new pay-per-use image models, available for all users through our new credit system and staking DIEM.

Nano Banana and the new Nano Banana Pro are now live, with increased prompt and token limits. Google's latest image model excels at distilling complex prompts into clear imagery and infographics Nano Banana Pro is truly impressive and we believe it to be a significant step in image models.

Seedream is also now available as a pay-per-use image model.

App

• Released a new, simple model switcher and a new action menu to all users.

• The model switcher now rearranges user tags and triggers settings/actions on hover.

• The action menu fixes a bug with the Video button on img2video models.

• Temporary chats are now live for all users, with the selection persisting across sessions and the state reflected in the chat input.

• Added a progress circle to display context usage and a circular progress indicator with a tooltip for character count.

• Added a confirmation dialog before video regeneration and consolidated confirmation button styling across the app.

• Fixed an issue where the "Create Account" link in the video notice was broken.

• Improved LaTeX rendering to allow more single-dollar-sign inline math correctly.

• Prevented HTML and meta tags from being rendered in chat messages.

• Treats HTML as plain text in user message markdown.

• Fixed a bug with the Support Bot repeating its answer and another with its response rendering.

• Fixed the Character Action Menu button to only open on the intended hover event.

• Updated copy on the delete character confirmation and added validation for chat IDs passed through the URL.

• Disabled the original NanoBanana (V1) model.

• Fixed message/avatar alignment in multi-modal.

• Added support for text parsing from native Word (DOCX) files.

• Added support for XLSX document text parsing using markdown-tables.

• Removed the "new" tag from the support bot.

Models

• GLM 4.6 was set as the default for all users, then reverted due to high output token volume, and later reapplied.

• Added a multi-modal flag to the Gemma model.

• Qwen3-235b is retiring from the in-app model selection, replaced by new Qwen235b API model IDs.

• Added private tags to LLM and image models for better filtering.

• Added anonymized tags to text models.

• Prompt Enhance model has been changed to GLM.

• Wan 2.5 is now the default video model.

• Added negative prompt support to the video settings modal for models that support it.

• Kimi was hidden as reasoning is not supported.

Video

• Set Wan 2.5 as the default video model.

• Changed video model badges layout to prevent the model name from wrapping.

• Added a dialog for credit confirmation before video regeneration.

API

• Added support for reasoning_content instead of <thinking> tags for reasoning models.

• Added mini app tags to the CMS at the request of Coinbase.

Fixes and Improvements

• Fixed an issue with double charging for pay-per-use image models.

• Fixed a missing icon for VEO Fast.

• Fixed an artifact bug and added a watermark to Seedream images.

• Fixed a table overflow-x issue.

• Improved stop/cancel behavior for multi-modal requests.

• Fixed an issue where deleting a folder/character would not delete the conversations within it.

• Memoized the gauth client promise to avoid duplicate promises during await.

GLM 4.6 is now available for Pro users

Developed by Zhipu AI, this model benchmarks extremely high against both closed and open source models. It performs well in character chats and creative writing but mainly excels in tasks where you want a smarter model for analysis or structured problem solving.

Please note that GLM 4.6 is currently live without reasoning.

Web Scraping is Live in the app and API

You can now turn any URL into AI context on Venice Just include a URL in your prompt, and Venice will automatically scrape the page to include as context for your request

Full blog announcement: https://venice.ai/blog/web-scraping-live-on-venice-turn-any-url-into-ai-context

App

• New Library: We are inviting all users to use the all-new Library, a redesigned space to manage and view your generated content.

• Video regeneration: You can now change the model and video settings when re-generating videos, giving you more creative control.

• HEIF Image Support: You can now upload and use HEIF images in the Image Editor and Upscaler.

• Markdown Tables: Added support for GitHub-flavored markdown, including tables, for richer formatting in chat.

• URL Detection: The URL detection feature is now enabled for all users, automatically identifying and processing links in your prompts.

• Video & Image Variations Badge: Added a badge indicator on the chat submit button to show the count of video and image variations to be generated.

• UI/UX Polish:

  • Updated the state label for the web search toggle button for clarity.

  • Fixed an Infura connection error message.

  • Tweaked character escaping to correctly display bold text in edited messages.

  • Fixed text message alignment and edit message padding.

  • Resolved a Support Bot submit button issue on certain screen sizes.

  • Fixed missing action buttons below the last generated video variations.

• Video Preview: Added support for video in the photo preview on the chat interface.

• Mini App: Implemented auto-connect functionality within the mini app on Base.

• Mobile App: Updated the Android APK download URL to point to version 1.3.0.

• ASR on Mobile: Automatic Speech Recognition (ASR) is now available for mobile beta users.

Models

• Added GLM 4.6: GLM 4.6 is now available for Pro users.

• Added Dolphin Mistral 3.2 (Beta): The new Dolphin Mistral 3.2 model is now available in beta.

• Added Veo 3: By popular demand, we have brought back the Veo 3 video generation model.

• Removed Wan 2.5 Safety Check: The safety check on the Wan 2.5 model has been removed for a more open generation experience.

• Decommissioned Flux Models: We have decommissioned the older Flux generation models to streamline our offerings.

Features

• Memoria (Beta): Memoria is now in beta, providing advanced memory retention for longer conversations and more coherent character interactions.

• Character Share Links: Added a share link to the character menu, making it easy to share your custom characters with others.

Fixes & Improvements

• PDF Uploads: Resolved an issue that was blocking multilingual PDFs from being processed correctly.

• Image Upscaling: Improved image quality for upscaled outputs and forced PNG format to remove jpeg compression artifacts.

• Rate Limit Messages: Fixed incorrect rate limit messages for anonymous and free users across chat, image, and video generation.

• Credit Balances: User credit balances now update correctly after usage, DIEM stakes, and unstakes.

API

• Search-Based Pricing: Enabled usage-based pricing for web search for all users.

Thanks for your patience in-between release notes — the Venice team has been hard at work over the last month at our all-hands offsite preparing for Venice V2 and shipping our most requested feature to date, Venice Video.

Moving forward, release notes will move to a bi-weekly cadence.

Venice Video

Video generation is now live on Venice for all users.

You can create videos on Venice using both text-to-video and image-to-video generation. This release brings state-of-the-art video generation models to our platform including Sora 2 and Veo 3.1.

You can learn all about this offering here.

Venice Support Bot

• Launched AI-powered Venice Support Bot for instant, 24/7 assistance directly in the UI.

• Bot pulls real-time information from venice.ai/faqs to provide up-to-date answers to common questions.

• Users can escalate to create support tickets with context when additional help is needed beyond FAQ responses.

• Available in English, Spanish, and German.

• Accessible via bottom-right corner (desktop) or chat history drawer (mobile browser/PWA).

App

• Launched Image “Remix Mode” - This is like regenerate, but uses AI to modify the original prompt. This provides an avenue to explore image generation prompts in more depth.

• Added “Spotlight Search” to the UI. Press Command+K on a Mac or Control+K on Windows to open the conversation search.

• Add a toggle in the Preferences to control the behavior of the “Enter” key when editing a prompt.

Characters

• Venice has launched a “context summarizer” feature which should improve the LLMs understanding of important events and context in longer character conversations.

API

• Added 3 new models in “beta” for users to experiment with:

  • Hermes Llama 3.1 405b

  • Qwen 3 Next 80b

  • Qwen 3 Coder 480b

• Retired “Legacy Diem” (previously known as VCU). All inference through the API is now billed either through staked tokenized Diem or USD.

API Pricing Updates

We’ve rolled out pricing drops across the Venice API’s specialized chat models (input/output):

• Venice Small (qwen3-4b) → $0.05 / $0.15 (was $0.15 / $0.60)

• Venice Uncensored (venice-uncensored) → $0.20 / $0.90 (was $0.50 / $2.00

• Venice Large (qwen3-235b 2507 thinking and instruct) → $0.90 / $4.50 (was $1.50 / $6.00)

Build more, pay less. See the full pricing page: https://docs.venice.ai/overview/pricing

Qwen Image Performance

We have updated Qwen Image and fit a Lightning LoRA to it. Few changes:

• Generations should run in about 1/2 the time as plain QI

• As per recommendation from LoRA authors, we have pinned the CFG and Steps in the backend to 1 and 8 respectively.
  

App

• Fixed pinch to zoom when the photo viewer is open.

• Made additional improvements to LaTeX rendering used for mathematical equations.  

• Fixed copy action not properly including in-painted / edited images.

• Changed default for new API keys to “inference only” keys vs. “admin” keys.

• Migrated to Qwen Edit as the default image editing / in-painting model.

• Updated to latest version of Wallet Connect to improve wallet linking user experience.

• Updated execution timer to reflect all images when generating multiple variants.

• Improved rendering when printing text chats.

• Improved auto model routing to direct anime related prompt to anime optimized models.

Social Feed

• Updated social feed rendering to improve rendering of non-square aspect ratio photos.

• Added a Download item to post share menus so users can save post images.

• Added a “My Posts” tab to the social feed navigation.

API

• Added an optional strict boolean on tool parameters to enable stricter parameter validation for tool calls.

• Added model deprecation headers (x-venice-model-deprecation-warning and x-venice-model-deprecation-date) to API endpoints.

• Added photoUrl parameter to character data returned from the API.

• Added an API route to get details on a specific character. API docs have been updated.

• Added support for multiple image generations via the API in a single request via the variants parameter. 

Today’s cover art can be found on the Venice Social feed.

Web App Updates

• Enhanced chat search across conversation titles, message content, and attachment names.

Tokenized DIEM Launched on Wednesday, August 20th

Last Wednesday, Venice launched support for Tokenized DIEM, including a new token dashboard design. Full details of the launch can be found here: https://venice.ai/blog/7-days-to-diem

Upcoming API Model Retirements

Venice has been steadily adding new models, and as users adopt them, we begin phasing out older ones with lower usage. A few models are scheduled for retirement, each with a stronger recommended replacement. Deprecation warnings will appear in your API responses when using these models until the sunset dates listed below.

You can see all the details about our deprecation policy in our Deprecations docs. Please check this page to understand:

• Our lifecycle policy (when and why models get retired)

• How deprecation warnings work in API responses

• What happens after sunset dates

• The live Deprecation Tracker (always up to date)

Models scheduled for deprecation → Recommended replacements:

Removal date: Sep 22, 2025
deepseek-r1-671b → qwen3-235b
llama-3.1-405b → qwen3-235b
dolphin-2.9.2-qwen2-72b → venice-uncensored
qwen-2.5-vl → mistral-31-24b
qwen-2.5-qwq-32b → qwen3-235b (with disable_thinking=true)
qwen-2.5-coder-32b → qwen3-235b
deepseek-coder-v2-lite → qwen3-235b
pony-realism → lustify-sdxl
stable-diffusion-3.5 → qwen-image

Removal date: Oct 22, 2025
flux-dev → qwen-image
flux-dev-uncensored → lustify-sdxl

Calls to these models will keep working until sunset, but with a deprecation warning. After sunset, direct calls to these IDs won’t return results unless you're using them via traits. We highly recommend using traits (default_code, default_vision, default_reasoning, etc.) since they always map to supported models.

Today’s cover art can be found on the Venice Social feed.

Introducing the Venice Mobile App

Venice is the only AI app that keeps your conversations completely private. Get honest answers to any question, generate stunning images, and create without restrictions on your mobile device. See our announcement on X.

• Download the Venice App for iOS

• Download the Venice App for Android

DIEM Launch Scheduled for Wednesday, August 20th

After months of development and community feedback, we’ve locked in the final parameters and launch date for tokenized Diem – an important upgrade to VVV’s tokenomics and a transformative evolution of our AI inference capacity system.

Critical Launch Details

• Launch Date: Wednesday, August 20, 2025

• Base Mint Rate: 90

• Adjustment Power: 2

• Target Diem Supply: 38,000

• Venice Minting: 10,000 DIEM (Mint Rate = 93.34 after Venice mints)

• VVV Inflation Reduction: Effective as of Aug 20 (14M → 10M/year)

READ ALL IMPORTANT DETAILS here: https://venice.ai/blog/7-days-to-diem

New Pro Model: Anime WAI

Launched a new Anime focused model based on WAI for Pro Users. This is a specialized model trained specifically for high-quality anime-style image generation. It understands anime aesthetics, delivers clean lines and vibrant colors, and produces results that look professional.

New Model: Qwen Image

Launched Qwen Image model for all users. If you like the prompt adherence of Hidream, the versatility of Flux, and the realism of Venice SD35, we just rolled out a new model for you: Qwen Image.

App Updates

• Announced upcoming retirement of Flux and Flux Uncensored image models.

• Enforce Reverification for user sensitive actions (like deleting an account).

• Fixed a layering issue in the Character creation modal where the footer could overlap content or interactive elements. The stacking order has been adjusted so the footer no longer obscures underlying UI, improving visibility and clickability.

• Improved responsive behavior: when starting a character edit on small screens, the chat bar now auto-hides to maximize space for the editor.

We hope you’ve been enjoying your Summer and have found an opportunity to adventure out and explore with family and friends.

Here’s the latest from the Venice team covering the last few weeks of our development. Stay tuned in the coming weeks for a number of major new releases from our team.

Today’s cover art can be found on the Venice Social feed.

Models

• Upgraded Venice Large to Qwen 235B A22B 2507, the latest version of this premier model published by the team at Qwen. Venice now routes thinking and non thinking requests programmatically between the two optimized models. You can learn more about this update here.

• Increased context size on Venice Small (Qwen 3 4B) to 40k tokens to fully expose model’s available context.

App

• Launched “Auto Mode” as the default for new Venice users. Auto mode will route prompts to the best model to answer the user’s questions, removing the need for users to swap between image and text generation. You can learn more about this update here.

• Migrated model settings from the top right corner of the chat window into a drop down selector in the chat input. You can learn more about this update here.

• Added a new “Settings” menu to the left nav bar to unify all Venice settings into a single place.

• Added “thumbs up” / “thumbs down” rankings to results to allow Venice to better gauge model quality based on user feedback.

• Permit users to auto-generate tags when posting a new social feed image.

• Added a UI element to show when a user has custom system prompts enabled.

• Added support to lock image settings to a specific model.

• Improved Venice’s models knowledge of its own features and capabilities to permit the model to guide users on how to use Venice more effectively.

• Added the concept of current user time to the models to permit more relevant responses when dealing with time pertinent requests. 

• Combined the “Edit Image” and “Enhance Image” modals into a single user interface for simpler image edits.

• Added UI states to report when the local browser database is inaccessible due to resource constraints imposed by the user’s browser.

• Permit any image in the chat history to be regenerated — previously only the last image in the conversation list would permit regeneration.

• Updated the social feed notification drawer to mark all notifications as read when the drawer is viewed.

• Updated content violation interface on image generation to give more information on what happened and provide a mechanism to regenerate certain prompts on our default model.

• Improved bulk chat deletion UX to avoid accidental clicks during flow.

• Permit the deletion of image variants when generating more than 1 image at a time.

• Show an improved error message when users are trying to save a social feed username that’s already in use.

• Fixed a bug where users were unable to preview the original source image when image mode was enabled.

• Fixed a bug that was stripping EXIF data from WebP images when they were downloaded.

• Fixed a bug where when users created a new conversation, the type would be reset back to text. Instead we now carry over whatever the last conversation type they were previously using, or if in auto mode, that is selected.

Characters

• Updated the default character sorting order to require reviews on the latest characters displayed in the list.

• Added the ability for Character chats to support file uploads.

• Fixed a layering issue in the Character modal where the footer could overlap content or interactive elements. The stacking order has been adjusted so the footer no longer obscures underlying UI, improving visibility and clickability.

API

• Added an API Key Info route to return information about a specific API key. API docs have been updated.

• Added an API Docs page describing the Venice model suite.

• Create custom validations for API models to avoid infinite repetition from certain settings on certain models.

• Retired the Fluently image model from the API based on usage patterns.

• Fix API bug that was artificially limiting the max payload size of requests, preventing the full context from being utilized.