Changelog

Text models53

• DeepSeek V4.1 Flash 1017 Uncensored

  Uncensored abliterated variant of DeepSeek V4.1 Flash 1017

  All usersPrivate

• GPT-6.1 Terra

  Fast text model from OpenAI for visual reasoning and computer use

  All usersAnonymous

• Abliterated Large Flash

  Unrestricted abliterated text model from Abliteration AI, based on DeepSeek V4.1 Flash

  All usersAnonymous

• DeepSeek V4.1 Flash Uncensored

  Uncensored abliterated variant of DeepSeek V4.1 Flash

  All usersPrivate

• Kimi K3.1 Pro

  Flagship multimodal reasoning model from Moonshot AI with a 1M-token context window

  All usersAnonymous

• Kimi K3.1 Pro Highspeed

  Faster variant of Moonshot AI's Kimi K3.1 Pro for latency-sensitive workloads

  All usersAnonymous
• Show 47 more

Image models10

• Qwen Image 2.1 Turbo

  Faster image generation model from Alibaba Cloud

  All usersAnonymous

• Qwen Image 2.1 Turbo

  Faster image editing model from Alibaba Cloud

  All usersAnonymous

• Nano Banana 2.1

  Image generation model from Google

  All usersAnonymous

• Nano Banana 2.1

  Image editing model from Google

  All usersAnonymous

• Flux 3

  Image generation model from Black Forest Labs

  All usersAnonymous

• Flux 3

  Image editing model from Black Forest Labs

  All usersAnonymous
• Show 4 more

Video models6

• Kling 4.0

  Image-to-video generation model from Kuaishou

  All usersAnonymous

• Kling 4.0 R2V

  Reference-to-video variant of Kling 4.0 from Kuaishou

  All usersAnonymous

• Flux 3 Max

  Image-to-video model from Black Forest Labs, the Max tier of Flux 3

  All usersPrivate

• MiniMax H3 Max Recast

  Video recast model from MiniMax

  All usersPrivate

• Flux 3 Extend

  Video extension model from Black Forest Labs

  All usersAnonymous

• Grok Imagine 1.5 Lite

  Lightweight image-to-video model from xAI

  All usersPrivate

Audio models1

• ElevenLabs Voice Changer

  Voice conversion model from ElevenLabs that re-records speech while preserving delivery and emotion

  All usersAnonymous

Deprecated15

• Claude Opus 4.8 Fast→replaced byClaude Opus 5.5 Fast
• Hermes 3 Llama 3.1 405b→replaced byDeepSeek V4.1 Flash
• Claude Sonnet 4.5→replaced byClaude Sonnet 5.5
• GLM 4.7 Flash→replaced byGLM 5.3 Flash
• Inkling→replaced byMuse Spark Open
• Qwen 3.5 9B→replaced byQwen 3.6 35B A3B
• Grok Imagine High Quality→replaced byGrok Imagine 2.0
• Grok Imagine High Quality (SOTA)→replaced byGrok Imagine 2.0
• Background Remover
• ImagineArt 1.5 Pro→replaced byIdeogram V4
• Seedance 1.5 Pro→replaced bySeedance 2.0 Mini
• Seedance 2.5 US→replaced byFlux 3 Max
• Seedance 2.5 US R2V→replaced byMiniMax H3 Max R2V
• Veo 3 Fast→replaced byVeo 3.1 Fast
• Veo 3 Full Quality→replaced byVeo 3.1 Full Quality

Chat21

• Unified User-Action Card Style

  User-action cards such as tool approvals and connect prompts share a single redesigned layout with softer corners, status colors that reflect the answer, and a compact form on phones.

• Better Mind Search

  Mind search now uses full-text search instead of substring matching, returning more relevant agent discovery results.

• Free Tier Auto Default Model

  Free users on Pro-only chat or image models now run on the Auto default model instead of bypassing the Pro-only check.

• Classic Chat In More Menu

  Classic Chat now lives in the sidebar's More menu instead of the top level and pins itself automatically for frequent users.

• Lower Anonymous Chat Limit

  The daily chat limit for logged-out users drops from 10 to 8 messages per 24 hours, with free and other tiers unchanged.

• Shared Conversation 14-Day Expiry

  Unprotected shared conversations now stop opening after 14 days and their stored data is deleted daily.

• Show 15 more

Image, video and audio36

• Image Quote Sizing Details

  Image quotes now return the resolved sizing for each quoted request, so the approval card can show exactly what a paid generate or edit will produce.

• Media Browser For All Users

  The media browser for uploads and documents is now enabled for every user.

• Kling 4.0 Video Models

  Kling 4.0 text-to-video, image-to-video, and reference-to-video models are now available to all users in the app and API, with 3–30 second durations, 720p/1080p output, native audio, and multi-shot support.

• Kling V4 End Frame Support

  Video generation now supports an end frame together with multi-shot prompting for Kling v4, plus the raised R2V start frame limit.

• Edit Balance Holds

  Web /edit now places a balance hold for the estimated cost and settles it after generation instead of charging and refunding.

• Image Generation Balance Holds

  Image generation now places a temporary hold on the quoted credit amount before generating instead of charging first and refunding on failure.

• Show 30 more

Library and media1

• Media Browser

  A new media browser with Uploads and Generations tabs lets users browse generated media and uploads, with polished grid spacing, rounded corners, and a full-panel drop target for dragging files in.

  Beta

Account and settings3

• Provider Safety Warning Ladder

  Users who trigger provider safety warnings now receive escalating email notices and can face 24-hour or permanent restrictions on provider models.

• Anonymous Email Masking

  All accounts now use a masked anonymous email address routed through a relay instead of their real address.

• Signed-In Identity for Support Chat

  A new identity token endpoint lets the web app pass signed-in user identity to Intercom support chat.

Wallet and payments9

• Credit-Loss Acknowledgement On Downgrade

  Downgrading or cancelling now requires explicitly acknowledging how many bundled credits will be lost, with the change taking effect on a shown date.

• Token Dialog Transaction Locks

  Token dashboard dialogs now lock while a crypto transaction is pending, and phone sheets get a repositioned close button and tighter spacing.

• Subscription Renewal Banner

  Paid subscribers now see a renewal reminder banner above the chat when their subscription is due to lapse or renew within three days, with a direct way to renew.

• Multi-Month Crypto Subscriptions

  All users can now buy 3- and 6-month crypto subscription terms through the redesigned single-screen checkout on the BTCPay and Coinbase endpoints.

• H3 Max Promo Extended

  The 35% H3 Max user promo is extended through October 31.

• Pro Weekend Sale

  A 10% Pro discount auto-applied at Stripe, Coinbase, and BTCPay checkout during the weekend sale, which has now ended with crypto invoices returning to list price.

  Pro users
• Show 3 more

General6

• TBC Model Icon

  Mischief 1 now displays the official TBC logo as its model icon.

• Support Chat Moves To Intercom

  The in-app support widget is replaced with an authenticated Intercom Messenger that respects telemetry settings and keeps the same support entry points.

• Standardized Modal Close Buttons

  Close buttons on modals and drawers now share a consistent circular style across the app.

• Sidebar More Menu

  Characters, API, Token Dashboard, and Labs move into a sidebar More menu with a one-time popover explaining it, items auto-pin to the sidebar on first visit, and manual unpins are remembered.

• Standardized Tab Styles

  Tab controls across the app now use two consistent visual styles, line and segmented.

• Experiments And Waivers Removed

  Inference experiments are removed in favor of explicit product defaults, and paid tiers are now billed credits for all models without waivers.

Mobile app17

• Assistant Panel Polish

  The Android assistant panel gets an X close button, a combined action row, and a more responsive speaking orb animation.

• Android Default Assistant

  Android 12+ users can set Venice as the phone's default digital assistant, opening a voice panel from the side button or lock screen that starts a new chat.

• Native iOS Chat Popovers

  Chat popovers on iOS now use native menus for a more platform-consistent feel.

• Native iOS Image Menu

  Long-pressing images on iOS now opens the native context menu.

• Accurate Video Cost Quotes

  Video-to-video and upscale cost quotes now use the attached clip's actual duration, and the composer accepts video attachments up to 20MB.

• Native Markdown Rendering

  Enriched markdown in chat messages is now rendered with native components.

• Show 11 more

• Deprecate Claude Opus 4.8 Fast

  Fast mode on claude-opus-4-8 enters a 14-day deprecation window ahead of the upstream retirement, after which speed fast requests return an error.

• Adaptive Reasoning On GLM And DeepSeek

  GLM-5.3 and DeepSeek V4.1 Flash now accept adaptive reasoning effort.

• Grok Responses Support

  The grok-4.3 and grok-4.6 routes now support the responses endpoint, with per-model output token limits and reasoning effort levels.

• Selectable Voice Model WebSocket

  The conversation websocket accepts an optional model parameter so clients can choose between the xAI voice models, falling back to the default when the value is missing or unsupported.

• Agent And Favorites Search

  Agent search and favorite agent listing now share one search path, with updated sort order, ranking, and filtering. Existing nextCursor tokens stop working.

• Faster DeepSeek V4.1 Flash

  DeepSeek V4.1 Flash API requests are now served on a faster route.

• Show 51 more

• Auto Mode Model Fallback

  Classic Auto mode and agentic chat fall back to the default LLM when their free model is unavailable instead of running on an inactive model.

  Chat

• Display Math Renders On Android

  Display math formulas now render correctly in Android release builds instead of showing blank gaps.

  Mobile

• Voice Mode Self-Interruption Fix

  iOS voice mode no longer cuts itself off mid-reply by mistaking its own playback for user speech.

  Pro usersMobile

• Fix Responses Replay Item IDs

  Compatibility history replayed through the native Responses API now emits valid item IDs, so replays and compaction no longer fail upstream validation.

  API

• Assistant Lock Screen Handoff

  Messages sent from the Android assistant on the lock screen now reliably open Venice to the new chat after unlocking instead of being lost.

  Mobile

• Inline HTML Sandboxes Restored

  Completed HTML sandbox results render inline in chat again, and the workspace panel now tracks files by path.

  Chat
• Show 65 more