New Models

• Added GPT Image 2.5 Flare and GPT Image 2.5 Sunburst.

New Features

• Added Workplace Safety & Health tool for Plus and Pro users. Search WSHC guidance on potential hazards, precautions, and case studies, with source citations and relevant images.

Enhancements

• Updated Gemini Flash to 3.8 and Claude Fable to 5.1.

New Features

• Added research across sources. Chat can investigate complex questions across attached sources, gather supporting evidence, and identify conflicting or missing information.

Enhancements

• Model popularity now uses weekly usage counts. Fixed missing benchmark data and relative intelligence scores in model selection.

Bug Fixes

• Fixed duplicate prompt submission when double tapping the send button on mobile.
• Fixed network error message when stream is interrupted when PWA is suspended. On restore, attempt to resume generation response.

Enhancements

• Added High Input Token warning when prompt length exceed specified threshold.

General

• Signing up with disposable email is now blocked.

Chat

Agent

• Chat now works as a multi-step agent that can plan, use tools, inspect results, and continue working
• Added tool approval and user input requests
• Added stop generation and response steering
• Added persistent memory and semantic memory
• Added automatic context compaction for long conversations
• Web search can now bypass some website with captcha
• Added in-chat image generation
• Mermaid diagram can now be rendered in chat.
• Added CORENET-X search tool

Chat Features

• Model selection now displays more information
• Added response regeneration with multiple response versions
• Added speech to text for chat input
• Added text-to-speech (dictation) on responses.
• Added support for mention with @
• Added support for commands with /
• Now drag-and-drops support files in addition to images
• Added minimap for long chats on desktop
• Favourite are removed in favor for Pinned Chats
• Pin up to 20 chats to the top of sidebar for quick access

Files, Sources and Facets

• Added a workspace for chat to manage sources (files) and facets.
• Added support for PDF, DOCX, PPTX, spreadsheet, CSV, email, text, and web sources
• Able to reliably process large excel / csv files with thousands of rows and columns.
• Added native text extraction for compatible files.
• Added option to switch between auto, text or OCR extraction for files.
• Added Facets for creating structured content alongside a chat (Notes, Summaries, Quizzes, and FAQs)

Projects

• Projects are collaborative workspace.
• Mange members and permissions for projects.
• Project sources are now available to all project members
• Project chats are now private until shared

Skills

• Remove snippet in favour of skills, existing snippet will not be migrated to skills.
• Added support for skills. Unlike skills in coding agents such as Codex, BuiltChat skills are instruction-only. They guide the AI through repeatable tasks, such as writing emails in your preferred style, but cannot run code or interact with a computer or server.

---

Images

Workspace

• Image thread is now call image board
• Added image albums
• Images in board and albums can be arranged by drag-and-drop
• Sharing now supports individual images, albums, and boards

Generation

• Added more configuration options (size, quality, resolution, background, format and style)
• Added preconfigured styles
• Added custom styles
• Multiple image output are available on supported model

---

Customization

Appearance

• Added additional themes
• Added many options to tweak the interface

Personalization

• Custom instructions can be set up using guided or freeform setup
• Added memory - persistent and semantic

  Memory works best when it is small and useful. Avoid saving too many personal details, as this can make responses feel overly personalized.

---

General

Security and Privacy

• Everything is private by default. All of your content, including generated outputs are store

  Previously input and output images are publicly accessible though the link to the resources are impossible to guess. Chats and files are always private since the beginning.

Usage

• Replaced separate monthly chat and image allowances with one weekly usage budget
• Weekly usage now resets every Monday at 2 PM Singapore time
• Usages are based on AI usage throughout the app.
• Usage cost for different models can be viewed in the model selection menu.
• Added usage breakdowns and charts in Settings

Pricing

• There are 2 paid plans. Plus and Pro, $35 and $50 per month respectively.
• Existing users will pay the same price as before.
  • Lite plan users will be upgraded to Plus plan
  • Old Pro plan users will be upgraded to New Pro plan

Issues & Fixes

• Approximately 15% of chat input images from the previous version cannot be restored due to cleanup bug, affected images now appear as [unavailable image]