We’ve shipped nearly 200 updates in the last three weeks. From voice input and output, to a brand new way to see where AI models agree and disagree, to smarter context handling behind the scenes – this is one of our biggest update cycles yet. Here’s what’s new.
New Solution – the Adjudicator
The addition of the Adjudicator enables you to move from multi-AI disagreement to a recommended decision direction with one simple click.
The Adjudicator reads every hallucination, contradiction, correction, and blind spot across your AI conversation — then tells you exactly what to do about them.
Read more about the adjudicator at this link.
New Features
- Voice Input & Output — Speak your prompts with Speech-to-Text and listen to AI responses with Text-to-Speech. A floating audio player lets you auto-continue playback across multiple responses
- AI Power Selector — Pro and Frontier users can toggle between Full Power and Balanced mode to control model reasoning vs. cost per response
- Disagreement & Correction Index (DCI) — See exactly where AI models agree and disagree on each turn, available as a dedicated tab in the sidebar
- Adjudicator — Get an independent, detailed decision brief and proposed direction based on D&CI notes for more informed decisions and further chat continuation, with one-click export option
- Auto-Follow Chat — Toggle auto-scroll to always see the latest response as it streams in, with full visibility of the current bubble and the next AI’s activity indicator
- Document Export — Export Master Documents as DOCX or PDF directly from the app
- Gemini Native Web Fetch — Gemini can now read URLs you share in conversation at no extra cost
- GPT-5.4 — Now available for Frontier and Enterprise tier users
- LinkedIn Login — Sign in or sign up with your LinkedIn account
- Forgot Password — Reset your password directly from the login screen
- Spark Free Trial — Try Suprmind with no credit card required
- Tool Usage Transparency — See which tools each AI used (web search, file analysis, etc.) in a footer below every response
- Smart Selector — Let Suprmind automatically pick the best AI model tier for your question
- Changelog Notifications — A bell icon in the sidebar keeps you up to date on new features and improvements
- Auto-Recovery for Streaming — If a response stream drops mid-way, the system automatically reconnects and resumes from where it left off
Improvements
- Smarter Web Search — Web search is now available on all tiers including Spark, with citation URLs shown alongside every response
- Better Context Handling — Major upgrade to how conversation context is built and shared across AI models — token-based compression, dynamic history windows, and smarter summarization improve response quality in longer threads
- Improved Document Export Quality — Fixed table formatting, page breaks, character spacing, and color rendering across both PDF and DOCX exports
- Refined Default AI Order — Gemini now responds first for faster initial results
- Master Documents Auto-Save — Copying or downloading a master document automatically saves it to your project knowledge
- Better Error Messages — AI provider errors are now shown in plain language instead of technical codes
- Bot Protection — Added Cloudflare Turnstile to login and signup for improved security
- Polished Settings Page — Redesigned profile tab with a cleaner two-column layout and theme toggle
- PWA Support on iOS — Proper spacing for iPhone notch and Dynamic Island
- Wider Project Sidebar — More room to view your project context, knowledge, and Scribe insights
- Cleaner Adjudicator Cards — Simplified card design with left-border accents instead of colored backgrounds
- More Readable Scribe Notes — Bumped small fonts and simplified category icons for better readability
Bug Fixes
- Fixed @mentions incorrectly triggering on attached file content
- Fixed Claude occasionally returning an empty response after using too many tools
- Fixed GPT-5.2 getting stuck in a tool-use loop
- Fixed mic button sometimes not responding to tap-to-stop
- Fixed file attachment count showing incorrect number
- Fixed signup and checkout flow reliability (loading states, redirect timing, embed sizing)
- Fixed onboarding questionnaire responding slowly to clicks
- Fixed occasional app crash related to state synchronization
- Fixed auto-scroll not showing the full message bubble
- Fixed light theme inconsistencies across the app
- Fixed Gemini thinking mode configuration for more consistent responses
