What’s new
What’s changed
New features, improvements and fixes — as soon as they’re live.
November 10, 2026
Uploading photos and editing images further
- Photos with the wrong file extension (such as an iPhone photo in HEIC format named .jpg) are no longer rejected with “The file content does not match the file type”. The studio detects the real format and converts HEIC, AVIF and TIFF to JPG.
- In image mode, the most recently created image is edited further automatically. Above the input field you see “Editing: last image”; the cross creates a new image instead. Attach older images via “Template” as before.
- “Improve prompt” keeps running if you click elsewhere in the meantime. Questions and the suggestion wait in the chat until you come back.
November 9, 2026
Date and time on every message
- Every chat message now shows its date and time, e.g. “09.11.2026 07:48”, so you can see at a glance how old an answer is.
- Automations with “Only notify when there is something to report” now leave a short note in their chat on every run, even when nothing happened: “Nothing to report” and what was checked. You still only get an email when there is something to report.
November 8, 2026
Discount codes for partners
- Partners create their own discount codes in the partner area: up to their commission rate, at most 50 % off the monthly subscription, for 1 to 6 months, redeemable any number of times. Whoever redeems a code becomes the partner’s customer. The discount is deducted from the commission for that customer, after which the full commission applies again.
- If you redeemed a discount code, you see the discount when subscribing and it is deducted automatically.
- GPT Image 2.6 image models: fewer unnecessary rejections by OpenAI’s safety filter, especially when editing your own photos.
November 7, 2026
Text size of your choice and New chat at the top right
- New under Settings → Account: the text size in four steps from Small to Extra large. It applies throughout the studio, including buttons and menus, and is saved to your account, so it also works on your phone and in the app.
- Small labels are now at least as large as the rest of the small text — some were previously fixed at 10 or 11 pixels and hard to read on phones.
- On phones there is now a button for a new chat at the top right next to the question mark, without going through the menu.
- Lots of models in your selection? Every now and then the studio offers to keep only the most important ones — the strongest from each major provider, an affordable one for everyday use and a few image models. The rest is only hidden; “Don’t show again” turns the hint off.
- Fonts in the brand preview, style guide and application designs now come from our server instead of directly from Google. Your browser no longer contacts Google for them.
November 6, 2026
New image models: GPT Image 2.6, weather from a new source
- New image models: GPT Image 2.6 Flare for fast results and GPT Image 2.6 Sunburst for elaborate motifs. Both render text in images cleanly, can edit attached images precisely and run directly at OpenAI.
- Weather information in voice conversations and automations now comes from the Norwegian Meteorological Institute, worldwide and with rain amounts including times. Sunrise, sunset and the chance of rain are no longer included.
November 5, 2026
Automations: only notify when there is something to report
- New option “Only notify me when there is something to report”: on every run the automation checks the condition from your task — “if it rains”, “if there are new results”, “if the price drops below €100” — and only then sends a message. Otherwise it stays quiet; the run is listed with its reason under “Recent runs”.
- Emails are sent by the studio, not by the model: if “Result” is set to “Send by email” or “Chat + email”, the result goes to your account address. The automation no longer claims it cannot send emails, and the form points it out when your task asks for an email but the result is only stored in the chat.
- Weather in automations now comes from real forecast data instead of web search. With “Allow web search” the automation can also read a page you name in the task — for example to check it for changes every day.
November 4, 2026
New: Agent — tasks the studio completes for you
- A new building block in the sidebar: you describe a task in words, the agent plans, researches the web, operates websites, creates files, slides and spreadsheets, and asks when it needs you. You see the progress live, answer follow-up questions and receive the result files under ’My Files’.
- Pay-as-you-go billing via your credit with a fixed cap per task — the task stops at the cap. Activation per account by the administrator; not available in EU-only mode because the provider is based outside the EU.
- A task is a conversation: even after “Done” you can reply, ask for changes or have it continue. If the agent asks a question, it is highlighted above the progress, and its suggested replies appear as buttons.
- When you create a task, it now shows who runs it — today Manus. “Agent” is the area; further agents can be added as a choice in the same place.
November 3, 2026
Research: error from a source no longer aborts everything
- If a website returns an error during research (e.g. 404), the model now receives this as a tool error and continues working — previously this case was handled internally like a request abort. The tool card shows ’Source not reachable’.
November 2, 2026
Job search with more boards, projects on mobile
- Job search: alongside Google Jobs and jobs.ch, now also LinkedIn Jobs, Indeed, Monster and any other boards by domain (e.g. regional portals) — selectable per search, all sources run in parallel.
- Projects on mobile: a long chat title made the page wider than the screen, you could scroll horizontally and the invite form was squashed. Fixed.
October 28, 2026
Application: full-service area for CV, cover letter, design, job search and interview
- New ’Application’ area: your profile (CV imported from PDF, Word, photo or LinkedIn) and a file for each position with everything that goes with it — advert, company dossier, fit check, cover letter, tailored CV, appointments, notes, files and a chat that knows the entire file.
- Your own style with honest review: salutation, tone, length, taboos and wishes in words. The studio tells you what it advises against — you decide.
- Market and language per application: Switzerland, Germany, Austria or international; German, English, French or bilingual German + English.
- Designs: six layouts, recreate your own template, company look from the target company’s website, adjustments in words. Export as PDF and Word, also as folder.
- Job search: search profiles via Google Jobs and jobs.ch, each position rated against your profile, ’Apply’ creates the file. Daily or weekly with email for new matches.
- Interview: preparation folder with questions, answers from your CV, salary range and self-introduction — and practice interviews written or spoken with evaluation.
- Optimise LinkedIn profile: headlines, about text, bullet points for each position, skills and recruiter search terms as ready-made suggestions.
- Voice assistant: The conversation transcript now uses OpenAI’s new streaming model gpt-live-transcribe — more accurate transcripts in the overlay and in chat. The speech model itself (gpt-realtime-2.1) is already the latest.
- Share chat: The share button is now always in the bar above the chat, not just under ’More’. On mobile, the window opens cleanly from the bottom — before it stuck out on the left.
- Record meetings live: place device in the middle, start recording — the screen stays awake, the level shows whether all voices are coming through, you can pause, and the recording is saved continuously: if the browser or battery dies, you can finish it later. Processing starts immediately after you finish.
- Certificates and references: upload employment references, reference letters, diplomas and certificates in your profile — the studio decodes the reference code, uses quotes as evidence and attaches the documents to the application portfolio on request.
- Applications is available directly for all subscription and VIP accounts, including for sub-users. Home page, features page, FAQ and the app tour now include the new areas Marketing, Meetings, Podium and Applications.
- Slimmer sidebar: under Settings → Account → Sidebar you can hide areas you’re not currently using — you can switch them back on at any time.
- Uploads from iCloud Drive and OneDrive: longer wait with notice, instructions for each system and ’Try again’ when the browser can’t read a cloud file that hasn’t loaded yet.
October 27, 2026
Panel: workshop, roles, research, listening and questions
- New mode Workshop: Put in a text — one model is the author, the others edit from perspectives like target audience, clarity, brand voice or sales impact, each finding with quote and suggestion. The moderator keeps a criteria sheet, the author revises with a change log, and at the end there are before and after scores.
- Two-factor setup on mobile: instead of having to scan your own screen, a button opens the Authenticator app directly; the key can also be copied or typed, backup codes copied with one click. Admins can reset 2FA for a user if the phone is lost. At login there’s now ’Use backup code’, error messages distinguish between wrong code, too many attempts and lockout, and the key is in the standard format that Microsoft Authenticator also understands reliably.
- Voice assistant: The screen stays awake as long as the conversation is running — previously the phone would go to the lock screen and the microphone would get quieter. Also applies to listening in the podium.
- Questions from the audience: While the panel is running, you can throw in questions — they’ll come up in the next round, and all participants have to address them.
- Profiles at your brand: The three ideas from an idea forge now live under Marketing → Campaigns at your brand. Mark favourites — the campaign manager uses them as the basis for ads.
- Roles to choose from: each participant gets a role from the dropdown (Sceptic, Numbers person, Devil’s advocate, Lawyer, Strategist and more; in the Ideas lab, thinking methods like Provocateur or Trend scout) or your own in free text. An info icon on each field explains what it does.
- Research before the first round: the moderator searches the web in advance, all participants discuss on a shared factual basis with numbered sources — the sources appear in the result.
- Listen: each participant speaks with their own voice — live whilst the panel is running, or afterwards the entire discussion plus result. Use ’Read result aloud’. Requires voice mode.
- Follow-up questions: ask the finished panel questions — all participants answer in their role, the moderator puts it in context; question and answers also appear in the export.
- Auto-scroll like in chat: the view follows new contributions only whilst you’re at the bottom. If you scroll up to read, it stays put; an arrow takes you to the latest contribution.
- Credit and limits: At 80% usage you’ll receive an email, at the limit a second one — whether it’s your own monthly limit, allocated monthly budget or credit. Previously there was only the notification in the chat.
- Help: New answers about the Podium, customising your own Word and PowerPoint files, and what happens at the limit.
October 26, 2026
Sidebar and All chats
- Meetings: The meeting bot’s email address is now visible with a copy button under ’Send bot to meeting’ and in the meeting settings — add it as a participant in the calendar invitation and the bot will join automatically at the start time.
- All chats: The page couldn’t be scrolled on mobile — fixed.
- Sidebar: The buttons next to each chat (More, Project, Share, Delete) only appeared on hover on iPad and other touch devices from tablet width, where there’s no mouse — now always visible.
October 25, 2026
App in the App Store
- moewe AI studio is now available as an app for iPhone and iPad in the App Store (apps.apple.com/de/app/id6818512028): Subscription and credit optionally via Apple as in-app purchase or as before in the browser. Codes can only be redeemed in the browser. In Safari on iPhone and iPad, a notice about the app appears at the top.
October 25, 2026
Customise your own files
- Uploaded Word and PowerPoint files can now be changed in the chat without losing the design: ’Adapt my CV with these texts’ replaces the paragraphs directly in your file — fonts, colours, layout and images remain. Previously, the file was rebuilt.
October 23, 2026
Panel
- New in the sidebar: Podium — two to four language models discuss with moderation. They answer blind first, then in turns with objections, adoption only with justification and closing statement; a moderator model leads, recognises stalemate and writes the result with consensus, dissent table, opinion changes and recommendation.
- Idea forge for brands: Participants with different thinking methods deliver campaign ideas, steal and twist them from each other, a jury selects, the three best are developed into campaign briefs with objection and escalation. Also accessible via Marketing → Campaigns.
- Result as Word and PDF, ’Continue asking in chat’ transfers it to a chat. Cost cap $3 per podium, completion notification by email.
October 21, 2026
Meetings
- Scraping: After a scraper search, the scraping tools remain available in the follow-up question, even if the tick isn’t set again — previously the answer broke off with a tool error.
- Costs everywhere in euros: Chat, credit ring, usage, team and meetings now show amounts in euros gross for all accounts. In the meeting detail, it shows how much of that is the per-minute price.
- Meetings that the bot records via an inserted link are now named after the meeting in Teams instead of ’Microsoft Teams meeting’; if no title remains, the transcript takes its own.
- Meetings can be renamed (pencil next to the heading).
- Search above the meeting list: by title, date and summary.
October 20, 2026
Apple purchases
- Purchases and subscription renewals from the iOS app always go to the account that made the purchase — even if a different account is now logged in on the device.
- Multiple renewals reported all at once at app start no longer overwrite each other.
October 20, 2026
Store app
- iPhone/iPad app: Codes (credit, trial period, referral, activation) can now only be redeemed in the browser at chat.moewe-studio.de, not in the app — an App Store requirement. Redeemed codes naturally also apply in the app.
October 19, 2026
Deep research and scraping
- Deep research: If the question clearly wants a data list (addresses, contact details, email addresses, Excel, CSV…), the studio asks beforehand — deep research writes a report with sources and doesn’t collect datasets. With one click, it continues with scraping instead (if enabled), or deep research runs anyway.
- Scraping: For address and contact lists of companies and institutions, the agent now proceeds in stages — first Google Maps per region (name, address, phone, website), then contact details from the websites, finally an Excel file with sheets for companies and contacts. Each stage is confirmed individually with cost indication; LinkedIn only on request.
October 8, 2026
Meeting Agent
- Meeting bot: paste a Teams, Google Meet or Zoom link, or invite the bot mailbox to your appointment — the bot joins as a guest (name customisable), announces the recording notice, records, and the minutes are created automatically. With video if you like: slides and shared windows as stills with timestamps to jump into the video.
- Meeting manager: open to-dos and questions across all meetings (overdue ones marked), analysis with recurring themes, contradictions between meetings and overdue commitments, plus a briefing before your next appointment.
- New: Meetings (sidebar). Upload a recording from Teams, Zoom or phone, or record directly in the studio — transcript with speakers, minutes with summary, decisions, to-dos (who, by when), open questions and inconsistencies, export as Word and PDF, assign names to speakers, jump into the recording by timestamp.
- Query meetings: «What was that thing with Ralf?» — answer with timestamp from the transcript, even across all meetings. Email when the minutes are ready.
- Recording notice with three modes (automatically read aloud, verbal, none), custom text possible; audio is deleted after the retention period, transcript and minutes remain. Depending on settings, processing runs exclusively with EU models.
- Meeting bot: Teams asked ’Are you sure you don’t want audio or video?’ and the bot got stuck before joining — the dialogue is now confirmed, media permissions are granted in advance, and missing audio devices are reported immediately instead of failing silently.
- Admin → Cost review: The server share for the meeting bot is now a setting under Subscription & Pricing (default 0, because the bot runs on the existing server). Previously, a flat €15 per month was charged as soon as a bot had run — hence the minus in the Monday email.
- Credit ring in the chat header: The circle now shows how much of the last top-up remains. Previously it counted total usage since account creation, making it look half-empty for regular customers even when credit was nearly full.
- Meeting bot: If everyone simply hangs up instead of ending the meeting, the bot now leaves on its own — after five minutes without sound, after one minute of ’Waiting for other participants’, or when it’s the only one left. Idle time afterwards is neither recorded nor charged. If no one joins an invited meeting, it waits 15 minutes and leaves without recording and without cost.
- Sidebar: The account area at the bottom is now collapsed on all devices — one line with name and email, one click opens Settings, Credit, Team, Feedback and Sign out as a menu above. The chat list gets significantly more space, especially on iPad.
- Copy: Selected text from a response now arrives in Word, Mail and so on without background and font colour — previously, dark mode inserted black background with white text. Headings, lists, tables, bold and code are preserved.
- Transparency: Next time you open the studio, it will ask once for your consent that requests are sent to the AI providers of the chosen model — with a list of providers and a link to the privacy policy. Model selection now shows the provider and route for each model (direct, OpenRouter, Azure, Bedrock, Vertex). Status and withdrawal under Settings → Account → AI providers.
- Deep review: Excel files (.xlsx) and CSV can now be reviewed — worksheets are read as text with sheet names, the dossier captures numbers and terms as with documents. Previously it said ’no reviewable format’.
- Uploads: files from iCloud Drive or OneDrive that haven’t fully loaded no longer lead to a silent failure, but to a clear message.
- Uploads: Files from iCloud Drive or OneDrive that the Mac only downloads on access failed on the first attempt with ’could not be read’ — the browser now tries reading up to four times with pauses before reporting. Marketing → Brand: document upload now also accepts Excel and CSV.
- Chat list: Follow-up like in Outlook — right-click or via ’…’ to mark a chat as unread (blinking dot remains until you open it), flag it, and assign up to five custom colour labels. Filter chips above the list show only flagged or chats with a label. Read status, flag and labels apply across all devices.
- Admin → Cost review modules: The calculation deducted provider costs twice and counted revenue gross — making deep research falsely show as a loss. Now everything net and correct (result = revenue net − provider costs − fixed costs).
- Meeting agent: The bot’s announcement now sounds natural (OpenAI voice instead of robot voice when no Mistral voice is set — the announcement is a fixed sentence without meeting content). Participant names from the invitation and meeting interface flow into the minutes: speakers are assigned names, to-dos mention names instead of ’Speaker A’. When deleting a meeting, audio, video and slides are removed from disk.
- Web search: An explicit research request (’research …’, ’search the web …’) activates web search for that response even without the ’Web’ chip. When search is active, the model now knows this explicitly and no longer claims it lacks web search.
- Chat: ’Stop’ is now available immediately after sending, not just when the response starts running.
- Images: Below each generated image it shows which image model created it. Venice models like Qwen Image 3 Pro stopped delivering images (changed interface) — fixed.
- Sidebar: Width draggable at the right edge (double-click resets, remembered per device); truncated chat titles appear in full as tooltip on hover.
- Meetings: Failed processing can be restarted with ’Process again’ — audio and transcript are reused, nothing is charged twice. Cause of the first failure (admin account found no EU text model) fixed.
- Meeting bot: The end of a Teams meeting by the organiser wasn’t detected (bot kept recording) — now additionally: 60 seconds without visible call = meeting over. The display name was partially lost when typing in Teams (’ml studio’) — input is now checked and set directly if necessary.
- Meeting bot: Joining failed in the first real Teams meeting with ’__name is not defined’ (runtime helper missing in browser) — fixed.
- Meeting bot: Outlook invitations were silently discarded because the start time was read from the time zone block of the calendar file (year 1601) — now only the appointment itself counts; ignored invitations appear with reason in the diagnostic log.
- Meeting bot: Invitations by email from admin accounts were falsely ignored as ’not enabled’ — the admin role now counts as in the studio.
- Images: Automatic model selection now also recognises English prompts, quotation marks and the chat model’s indication of whether text should appear in the image — infographics, diagrams, logos and posters thus always go to a model that renders text cleanly (Nano Banana Pro, GPT Image), no longer to the cheapest. Without indication, a solid all-round model is suggested; the selection card shows ’Text ✓’ per model and always includes the strong models.
- Chat list: Chats with a new, not yet seen response show a blinking dot and highlighted title — handy when several chats are running in parallel tabs. The list updates itself every 15 seconds.
- Voice call: Web search only returned sentence fragments for factual questions (like holiday dates) because the search model used up its response budget thinking — now fixed; the assistant now passes a question that quick search can’t answer directly to the studio instead of searching repeatedly.
October 5, 2026
Campaign manager with Google Ads connection
- Marketing → Ads: Connect your own Google Ads account. The studio creates the planned campaign paused directly in the account (budget, locations, negative list, ad groups, keywords, ads, sitelinks) — you review and activate it in Google Ads. Google bills the ad spend directly to you as usual.
- Marketing package and campaign manager as modules: depending on plan, included at no extra cost for subscription and VIP accounts (as it’s currently set) or bookable individually — then pro-rated from booking day with subscription invoice, cancellation any time with usage until the end of the paid period. Model costs run via credit as always.
- Reports & optimisation: Load performance data from the last 7, 14 or 30 days directly from the Google Ads account (campaign, ad groups, keywords, search terms, ads) instead of exporting CSV. Recommendations with actions — pause, budget, bid, negative keyword — you apply with a click; each change confirmed individually.
October 2, 2026
Larger on mobile and tablet
- Knowledge base: search across all guides and FAQs, category bar for jumping, ’New’ badge for fresh guides; new Marketing category with guide to the Marketing Package.
- Files: When saving from the PDF view, the file is now named after the document instead of an identifier; in the Mac app, Word and Excel files download directly to the Downloads folder, on iPhone and iPad the button in the view is now called ’Share’.
- New: Campaign Manager (Marketing → Adverts). Adverts on Google, Microsoft, Meta, LinkedIn and TikTok from scratch: media plan, keyword research with negative list, ready-made ad copy for each channel, tracking links, import file for Google Ads Editor, Excel workbook and a step-by-step guide to a live campaign — including setting up the ad account and conversion tracking. Afterwards: upload performance report, get concrete optimisations.
- New: marketing package for €4.99 per month (subscription and VIP accounts, Settings → Credit or sidebar ’Marketing’): unlocks brand with style guide, campaign agent and content creator; campaign stations at fixed price from credit.
- New: brand (design system) in the ’Marketing’ area (unlocked by admin or via marketing package). Store logo, colours, fonts, tone, image style and target audience — or have it created in four questions (three directions to choose from) or derive from website and materials (style guide, presentation, brochure). One click generates the digital style guide as page and PDF. Selectable in chat as ’Brand’ chip: answers, images, presentations and documents then follow the brand. Foundation for the upcoming campaign agent and content creator.
- Voice conversation: Just say ’Create an Excel with…’, ’Turn that into a Word document’, ’Generate an image of…’ or ’Create a presentation’ — the studio handles it in the background, the file or image appears in the chat, and the assistant lets you know. It also passes demanding questions (calculations, analyses, facts, translations) to the powerful text model and reads the answer aloud; the detailed version is in the chat.
- Voice conversation: weather questions are now answered by the assistant with real forecast data (min/max, rain, wind, progression throughout the day) instead of vague search results; the ring stays round when the transcript gets longer.
- Voice conversation: Audio now runs without intermediate processing in the browser (cause of crackling and dropouts), speaker echo no longer interrupts the AI, and you’ll see the conversation cost in the top right; in the chat, the cost is shown below each spoken response.
- New: Deep check (More → Deep check). Have an entire book, contract, concept or thesis checked completely: spelling and grammar, contradictions between chapters (characters, numbers, timeline), structure, clarity, tone, formalities — with optional fact-check on the web. Report in chat and as PDF/Word, plus a corrected version as Word file with track changes. Runs in the background, email when complete.
- Installed app: Downloads (Excel, Word, PDF, images) now open the share sheet instead of the preview without a back button — the chat stays put.
- New: deep research (More → Deep research). Like ’Deep Research’: research plan, up to 30 web searches, a dozen pages read in full, gap check, report with sources for each statement. Runs in the background, 4 to 10 minutes, notification by email.
- Appearance can be selected: Settings → Account → Light, Dark or same as system.
- When reading whilst the response is still running, the view no longer jumps down. If you scroll up, it stays put; an arrow takes you back to the end.
- On mobile, text next to selection fields was squeezed into narrow columns (Settings → Models) — fixed, checked the entire app.
- Voice conversation: fixed crackling when the AI speaks (the animation was interfering with the audio stream); the first word of a question is no longer cut off.
- For operations: The exchange rate for credit now follows the ECB rate daily.
- On touch devices, text and buttons are now noticeably larger: chat text 17 px like on iOS, list rows taller, buttons finger-sized (42 px). Applies to mobile and tablet, including iPad in landscape. On desktop, everything stays as usual.
October 1, 2026
Project knowledge without file limit, memory everywhere
- Project knowledge: The five-file limit is gone. There’s now a storage limit per project (shown under ’Files’). With lots of text, the studio automatically extracts the relevant sections for each question — the cost per message stays low, no matter how many files are in the project.
- PDF export: Wide tables with many columns now fit on the page, column widths adapt to the content.
- Assistants: Knowledge files without file limit, with the same knowledge search as in projects.
- Attachments: Text files (.txt, .md) can now be attached in chat, in projects and with assistants. If a file type is not supported, the filename appears in the message.
- Fixed: The Claude models ’Bedrock EU’ returned ’Access not allowed’ — they’ve been removed from the menu until the provider grants access. Models that a provider denies will automatically disappear from the menu in future.
- Sidebar: Pen icon next to a project starts a new chat directly in that project.
- The voice conversation can now look things up on the web: "What’s the weather like today?" — it says "I’ll check", searches using your saved location and reads out the forecast.
- In auto mode, web search switches on automatically for current information (weather, news, prices, opening hours); saved facts like your location are used without asking.
- Automations now know your memory and profile — a morning briefing knows where you live.
- Fixed: ’View’ for Word and PDF documents showed ’Connection refused’ — the viewer wasn’t allowed to embed its own file.
- Admin: user list with filters by account type, partner and voucher; codes page with ’Delete all redeemed codes’ and ’Delete expired codes’.
- Partners can receive a share of their customers’ credit purchases if desired (admin enables this per partner).
September 30, 2026
Voice conversation, chat search, all chats, mobile view
- The studio can be installed as an app: on iPhone and iPad via Share → ’Add to Home Screen’, on Android, Mac and Windows via ’Install’ in the browser — full screen, own icon, offline notice.
- Trial codes are now also available for just a few days, for example 7 days free, not only in whole months.
- Zoom images in full view: mouse wheel, two fingers, double-click or the + and − buttons, up to 800%; when zoomed, you can move the image.
- Search chats: the search field above the chat list finds titles and message text. The list is grouped by projects and time (Today, Yesterday, last 7 days …).
- New ’All chats’ page: all conversations with search, project filter, sorting, multi-select to delete or assign projects.
- Voice conversation redesigned: microphone in the orb with glowing particle ring that reacts to your voice and the AI’s; large mute button (M key) for noisy rooms, display of Listening/Thinking/Speaking, conversation duration, «Reconnect» for disruptions.
- AI voice selectable: Settings → «Voice in conversation», ten voices with descriptions.
- New guide ’Which model for what?’ in Help: the strengths of Claude, GPT, Gemini, Grok, Mistral and the image models at a glance; FAQ expanded with chat search, sending, zoom, voice and model selection.
- On mobile: the Help button no longer sits over the close X of dialogues; the switch explanation (ⓘ) is now in the Help menu instead of alone in the chip row.
- Models: «Get suggested selection» in settings — the AI checks the current catalogue and hides everything except the currently best models, at least one each from Claude, GPT, Grok, Gemini and Seedream. Ticking and unticking works as usual afterwards.
- Voice conversation more stable: speaker echo and background noise no longer interrupt the response as easily (less crackling and dropouts).
- Enter now creates a line break, send via the arrow (or Ctrl/Cmd+Enter) — no more accidental sending.
- On mobile: input field across full width with buttons below, more compact sidebar with twice as much space for chats, image buttons as icons.
September 29, 2026
View documents, multi-image tidied up, partner for all accounts
- Larger attachments: Word, Excel, PowerPoint and CSV up to 50 MB, PDF and images up to 20 MB (previously 15 MB for everything).
- For operations: warning email when credit at an AI provider is running low.
- Voice conversation switched to GPT Realtime 2.1: understands better, speaks more naturally, can be interrupted; noise suppression for the microphone.
- Dictation: new transcription model, reliable in Safari too, clear message when microphone is missing or in use.
- PDFs, Word documents and design presentations can be viewed directly in the app — ’View’ on the document card, no download needed.
- Multi-image is now a pill in the toggle row in image mode; the options only appear when it’s on.
- Template and mark now also work with the 18+ image models.
- Partner programme: Partner is its own status, open to any main account; note on the pricing page.
- Prompt help in image mode starts with ’Suggestion’, so you see the improved prompt before generating.
- User management: expanding shows an overview, editing happens on a separate page per account.
- New: changelog, help with guides, error reporting with ’Report problem’ in chat.
September 25, 2026
Mark images, clearer projects, partner programme
- Mark images: paint areas red, green, blue or yellow and say what should happen there — the rest stays unchanged.
- Clicking images opens the large view, uploaded and generated ones.
- Multi-image: up to 4 models × 4 images, also from ’Auto’ via More → Multi-image; templates work there too.
- ’Improve prompt’ as a magic wand directly in the input field.
- Image tour: starts with the first image mode, anytime under Help.
- Project page redesigned: name directly editable, chats on the left, knowledge with auto-save and file chips, members on the right. Overview as tiles with last activity.
- Add chat to a project via folder icon — in the chat list, in the input field and on the chip at the top.
- Cost, copy, read aloud, PDF and Word under each answer always visible, even in Safari. Read aloud with more natural voice and stop button.
- Model selection with search, filters and groups in settings, team and admin.
- Partner programme: customers with a subscription can become partners and earn commission on their referred customers’ subscriptions.
- Error messages in chat now have ’Report problem’ — one click to reach us.
- Logo adapts to light and dark mode.
September 24, 2026
Child profile, new website, EU models
- Child profile with five age levels from 4 to 17: every message, every answer and every image request is checked, parents stay informed and can get an email for every alert.
- New website with subpages Features, Teams, Families, Safety and Pricing — in German, English and French.
- GPT-6 Sol, GPT-6 Luna and GPT-5.5 now run in the EU data zone (Azure), Claude Sonnet 4.6 and Opus 4.6 via AWS Frankfurt.
- Assistants: choose main task, ’Auto’ picks the right model.
- Formal or informal address in settings.
- Product suggestions with image, price and link when asking about products.
- Knowledge files in projects and assistants also as Word, Excel, PowerPoint and CSV.
September 23, 2026
Safeguard, three languages, memory
- Safeguard: IBAN, cards, tax ID, email, phone, addresses, names and passwords are masked before sending — the AI model only sees placeholders.
- Interface and emails in German, English and French; browser language is detected once.
- Memory with profile and automatic notes, opt-in, viewable and deletable anytime.
- My files: everything generated by month, with filter, search, download and delete.
- Control bar tidied up: only model, project, web, thinking depth and Safeguard directly visible, everything less common under ’More’.
Ready? Up and running in two minutes.
Create an account, choose a plan, load credits — and start working with all models right away.
Start now