MF0.ai updates

✅ GPT-6 Astra has landed — Open's newest flagship is in the ChatGPT list. Astra is built for the hardest end-to-end work: complex reasoning, coding, research and document creation, with a 1M-token memory of the conversation. It's an opt-in premium choice — the regular stays put, and the price is always shown up front in Pricing and in every answer's cost breakdown.

✅ Grok 4.7 is here — xAI's new flagship is now the default Grok in the model list: smarter answers, 500K-token memory of the conversation, configurable thinking depth. Previous Grok models stay available in Settings.

✅ Kimi learns web search — pick Kimi in "+" → Web search, and your question is first checked against the live web. The answer comes back grounded in freshly fetched pages, with links to the sources used.

✅ Mobile money for Africa — top up your balance with M-Pesa (Kenya), MTN, Airtel, Orange, Moov, Vodacom, Zam and more across 13 countries. Open the new Africa section in the top-up window, pick your operator, enter your phone number and approve the PIN prompt on your phone — the balance updates the moment the payment lands.

✅ More of Europe: EPS, Bancontact and Klarna — the Europe section of the top-up window now covers Poland (BLIK), Austria (EPS), Belgium (Bancontact) and the whole continent with Klarna. Pick your way, approve the payment on your bank's page — you're right back with the balance updated.

✅ Korean wallets in the new Asia section — top up with Naver Pay, Kakao Pay, Samsung Pay or PAYCO. Choose one in the Asia section, approve in your wallet — the balance updates the the payment clears.

✅ The interface speaks Korean — the whole product is now available in Korean: chat, settings, pricing and payments. That makes 18 interface languages in total.

🌟 AI agents are here — multi-step pipelines that work while you're away — give one task, and a whole crew of AI models handles it in sequence: a researcher digs in, an analyst structures it, a writer turns it into a polished text, a critic and an editor refine it further. Pick a ready-made recipe (research report, fact-check, debate, competitors table, rewrite-with-critique and more) or build your own chain of 2–8 stages — each with its role, its model and its instruction. Launch it and go about your day: the run works in the background, a live progress bar shows every stage, and the text lands in your chat history with a per-stage cost breakdown. Up to three pipelines in parallel, a stop button and a retry-from-the-failed-stage included. For now agents work with text models only — images and video join the crew later.

✅ GPT Image 2.5 is here — the native ChatGPT image models get two new members: Flare, faster and better than GPT Image 2 and now the default, and Sunburst, a premium tier for precise edits. The oldest models (GPT Image 1 / 1.5) are retired from the picker. And the quality picker gains two new top levels on the 2.5 models — Extra and Max — for the sharpest possible output, with the price always shown up front.

🌟 Six new languages — the interface now speaks H, Bengali, Indonesian, Vietnamese, Swahili and Tagalog, on top of the eleven it already had. Everything is translated: chat, settings, pricing, payments, the whole product — 17 languages in total.

✅ Google Pay and Apple Pay — topping up is now one tap when your wallet is set up: pick it in the payment list, confirm in your device's native, and the balance updates the moment the payment clears. If the wallet can't load in your browser, the window says exactly where it works.

✅ Pay the local way: Pix and BLIK — in Brazil you can pay with Pix: pick your, see the exact total in reais up front, scan the QR code in your banking app — done. In Poland it's BLIK: enter the 6-digit code from your bank app and approve. Whatever you picked in dollars is exactly what lands on your balance.

✅ Proper names for your chats — no more titled with a chopped-off piece of your first sentence (thanks for the report!). Every new chat now gets a real short title a few seconds after it starts, named quietly in the background — including chats created inside folders, which never got smart titles at all.

Woohoo! 🥳 We just hit a fantastic data milestone: 4,000 users on the project!

Here are the exact metrics that make this so exciting:
▪️ Total User Base: 4,000
▪️ Active Core Retention: 263 highly engaged, constant users
▪️ Media Support / Ad Spend: $500

This means our growth vector is 99% organic. We are scaling purely through unadulterated word-of-mouth and community traction, without external marketing boosts. The retention data speaks for itself — thank you to everyone fueling this organic machine!

✅ Memory Tree from big files — drop a large document or a long list into My intro, and the Memory Tree builds the full structure from it: the whole file is read to the end, nothing is cut off, and the facts are filed exactly the way you describe — what becomes a node and what goes into each description.

✅ Several photos in one video — Image → Video now takes multiple reference photos on MiniMax H3, Kling 3, Seedance 2.5 and Wan 3.0 / 3.0 Prime: every photo you attach guides the result (mention them in the prompt as Image 1, Image 2) — a person, a place and a style can meet in one clip. Before, only the first photo was used.

✅ Answers as Word documents — every text answer now has a ⬇️ DOCX button in its footer: one click downloads the reply as a ready Word document with real styles — headings, bulleted and numbered lists, tables, code blocks, links. No more copy-pasting into Word.

✅ Free joke avatar, no sign-up needed — the login page now opens with a freebie: upload any photo and the AI redraws it as a deliberately ugly, hilarious doodle character — crayon outlines, messy shading, silly scribbles and all. One tap on the share icon copies a link to show your friends, and anyone who joins MF0 from your link gets a $1 welcome bonus. Generations are free and unlimited — spread the doodles.

✅ Aspect ratios that match the model — the image aspect picker now offers exactly what the selected model can produce, nothing invented: every model has its own list, and 3:2 and 2:3 are finally there on every model that supports them. Switch models and the list snaps to the new one's capabilities; a ratio you set stays whenever the model allows it.

✅ Your picked model no longer switches itself — a whole family of glitches is gone: the image model used to change quietly after you attached a reference photo, or kept the previous model's settings and generated with the wrong one. What you choose in the dropdown is what generates now, always with its own proper parameters.

✅ Card top-up fixed on every screen — an extra sign-up box inside the payment form used to cover the Pay button, making it impossible to pay on phones and desktop alike. The card form is now short and clean, and the Pay button is always within reach.

✅ Contact search confirms the start — launching an outreach round now shows a small confirmation with the search quality and the contact count, so you know the background run has actually begun. Close it with OK or Esc and carry on — the results land in My rounds.

✅ Contact search is live for everyone — describe who you need in plain words (who, and where) and the app goes out and finds the companies themselves, with their public business contacts: email, website, LinkedIn, Instagram, YouTube, TikTok. A finished round expands into a table you can export (CSV / JSON / XML / TXT), any round can be continued to look only for new contacts, and Analytics itemises what each model in the pipeline cost you. Worth knowing before you start: a round can take up to an hour but runs in the background — close the page, the result waits for you — and you are charged for the real search work, not for a promised number of contacts. The estimate price is always shown before launch.

✅ A dossier and a ready outreach email for every contact — switch on the profile toggle and each company you find gets a detailed dossier, cross-checked against the pages the search actually fetched and written in any of the 11 interface languages: click the company name to open it. Add the message toggle and every contact also gets its own outreach text for email in English, grounded in that dossier and in your brief — open it from the Message link and copy it in one click.

✅ Folders work the same on your phone — tapping a folder in the Dialogs sheet opens it right there, so you can walk down a nested tree without the sheet closing on you. It closes when you actually go somewhere: tapping a chat, or picking New chat in a folder's ≡ menu — which starts the chat inside that folder.

✅ GLM steps out of Programming mode — GLM now sits in the regular chat model list as well, right after Kimi, and the GLM 5.3 ↔️ GLM 5.3 Flash choice lives in Settings → AI providers like every other model. The model menu is tidier too: the main providers and AI opinion come first, then Seed, Kimi and GLM after a divider, with Settings last.

✅ Folders inside folders — every folder's ≡ menu now has Create folder, so you can build a tree of any depth to organize your chats. Subfolders sit above the chats inside a folder, each with its own New chat / Rename / Delete actions, and Move to… shows the whole tree, so a subfolder is a valid destination like any other. Keep as many branches open at once as you like, at any depth; closing a folder hides everything inside it, and deleting one clears its whole branch after a clear warning.

✅ Folders work the same on your phone — tapping a folder in the Dialogs sheet opens it right there, so you can walk down a nested tree without the sheet closing on you. It closes when you actually go somewhere: tapping a chat, or picking New chat in a folder's ≡ menu — which starts the chat inside that folder.

✅ Seedance 2.5 joins Image → Video — a new model for animating a photo, with durations from 4 up to 30 seconds (4, 5, 6, 8, 10, 15, 20, 30) — the same 30-second ceiling Wan 3.0 and Wan 3.0 Prime already offer. It is the premium choice in that menu: noticeably more per second than the rest, so a long clip is a real spend. As always the per-second rate is in Pricing and the cost of your clip is shown before you run it.

✅ The app tells you when a big update is out — if a major MF0.ai update ships while your tab stays open, a popup in your language names the new version, and one click on OK refreshes you onto it. Only for real releases: the constant small fixes never interrupt you.

✅ Around the app — after reusing an older request (the ↻ button), the composer's model pill always shows the restored model; and the rename-conversation window now speaks all 11 interface languages.

✅ Longer videos from a photo — Wan 3.0 and Wan 3.0 Prime join Image → Video, and they take clips of up to 30 seconds — 5, 10, 15, 20 or 30. MiniMax H3 now offers 15 seconds too. Wan 2.7 stays right where it was, so nothing you were using has moved. As always the price is per second and you see it before you run.

✅ Two new coding models: GLM 5.3 and GLM 5.3 Flash — both join Programming mode alongside Kimi and the Nemotron models, billed per use like the rest of that group. Flash is the light one and is currently on a promotional rate from its maker — the Pricing panel and the "ⓘ" popup always show the rate you are actually billed.

✅ A roomier "+" menu — the attach menu now matches the width and spacing of the profile popup, so the mode names have space to breathe instead of being packed into a narrow column.

✅ GPT-5.6 Sol costs less to run — the strongest ChatGPT model in the picker dropped to $4 per 1M input tokens and $20 per 1M output: 20% off input, 33% off output, straight through to you. Its maker calls this a promotional rate held at least until 21 November — the Pricing panel always shows what you are actually billed, so if it moves, you will see it there the same day.

✅ Formulas look like formulas — LaTeX in a reply is typeset properly: $$…$$, \[…\] and inline $…$ all render as real mathematics, in chat and on shared pages alike. Ask for a derivation, a matrix or a bit of physics and you get something you can read. Prices in dollars stay plain text, as you would expect.

✅ AI settings save themselves — the Save button is gone from AI settings: pick a model or a preset and it is stored right away. Nothing to remember, nothing to lose by navigating away.

✅ Image settings follow the model you pick — choose a different image model and Ratio, Resolution and Quality snap to what that model actually supports; pick the same one again and your tuning stays as you set it, 4K included.

✅ Around the app — the model dropdown now has a ⚙️ Settings entry at the bottom that takes you straight to AI settings, and on a phone the Settings page gets a round ✕ close button within thumb reach at the bottom-left.

✅ Your Memory Tree can change its mind — when a fact about you is replaced rather than added to — you moved city, changed jobs — the tree now retires the old line instead of keeping both forever. Answers stop mixing your past with your present.

✅ Web search and Deep research moved to a new engine — the Perplexity-backed modes now run on the provider's current API, and we bill exactly what the provider reports for the request, search fees included. On the same research question we measured, that came out about 27% cheaper than before.

✅ Voice reads the entire reply — long answers are no longer cut short when you ask to hear them: the text is synthesized in pieces and played back as one seamless recording. In our test a 6,400-character answer became a single 7-minute audio.

✅ A gold star for what you haven't seen — a shimmering star on your profile icon means you unlocked an achievement since you last looked. Open the profile and it goes quiet again.

✅ App updates don't interrupt what's running — a reply that is streaming, or a Fusion or Ultra run in progress, finishes even if we ship an update at that exact moment. Generations already survived updates; now the text side does too.

✅ Memory Tree and Digital Brain take turns — only the view you are actually looking at keeps rendering, so switching between the two stays smooth even with a thousand-node tree.

✅ Clearer wording in friend chats — the chat menu now says Set password and Remove password, which is what the action does.

✅ Text → Video is here — a new plugin mode in the "+" menu: write a prompt, get a video. Seven models to pick from (Kling 3, Veo 3.1, Seedance 2.5, Flux 3, MiniMax H3, Grok Imagine Video, Gemini Omni Flash), each with its own duration, aspect ratio, resolution and sound settings — and as always, no subscription: you see the price before you run it.

✅ The model lineup just got a refresh — new chat defaults: GPT-5.6, Claude Sonnet 5 and Grok 4.5, with Claude Opus 5, Fable 5 and Gemini 3.6 Flash joining the picker. New on the media side: Flux 2 Max and Seedream 5 Pro for images, Flux 3 and MiniMax H3 for video. Prices shown are the prices billed, verified against the makers' own pages.

✅ Claude Sonnet 5 is a third cheaper — the model maker made its reduced launch pricing permanent, and we pass it through as-is: Sonnet 5 answers now cost ~33% less.

✅ A long answer arrives whole — if the connection drops or the model's servers go quiet mid-reply, the answer picks up from the exact word it left off, in the same bubble. When it truly can't continue, it says so plainly instead of leaving you with half an answer that looks finished.

✅ Stop stays with a heavy run from start to finish — in Fusion and Ultra you can stop at any point of the run, including after a page reload, and you pay for the steps that actually ran.

✅ Incognito leaves nothing behind — clear an Incognito chat and the pictures and videos you made in it leave our servers with it; close the tab instead and they are swept within 24 hours.

✅ Video → Video checks your clip before you spend anything — Kling 3 takes clips up to 10 seconds and frames between 512 and 2160 pixels. Attach something outside that and the app tells you at once, offering to trim the clip to the first 10 seconds right there — in your language, before a single cent is spent.

✅ Straight to the answer — replies open with the answer instead of a greeting: one short welcome in a brand-new chat, then business, in every mode including Incognito.

✅ Bug reports now get an answer — when the issue you reported is fixed, you'll get a short email in your language saying it's resolved, with your original report quoted. No more shouting into the void.

✅ The model lineup just got a refresh — new chat defaults: GPT-5.6, Claude Sonnet 5 and Grok 4.5, with Claude Opus 5, Fable 5 and Gemini 3.6 Flash joining the picker. New on the media side: Flux 2 Max and Seedream 5 Pro for images, Flux 3 and MiniMax H3 for video. Prices shown are the prices billed, verified against the makers' own pages.

✅ Claude Sonnet 5 is a third cheaper — the model maker made its reduced launch pricing permanent, and we pass it through as-is: Sonnet 5 answers now cost ~33% less.

✅ Replies now finish even when the connection hiccups — if the stream drops mid-answer, the reply automatically resumes from the exact spot in the same bubble, so a long answer arrives whole instead of stopping mid-word.

✅ No more automatic greetings — replies no longer open with a "Hello!" or "Great to see you!" every single time; one short welcome in a brand-new chat, then straight to the answer, in every mode including Incognito.

✅ Bug reports now get an answer — when the issue you reported is fixed, you'll get a short email in your language saying it's resolved, with your original report quoted. No more shouting into the void.

⚠️⚠️⚠️

🕶 Incognito runs each model at its most permissive setting — generations started from Incognito now ask every model for the least restrictive moderation it actually offers (some have no off-switch, so this is their floor, not none). And as always in Incognito, nothing is saved.

💰 A paid result is always yours to pick up — even if a heavy generation pushed your balance below zero, you can still check the task's status, collect the finished result and cancel a running task; only starting new paid work waits for a top-up.

📥 Claude import brings the whole archive — importing a Claude conversation export now adds every conversation in the file to your chosen theme, each as its own chat.

🔊 Voice replies are instant on replay — a generated voice reply is now kept, so listening again starts immediately instead of regenerating the audio from scratch.

🧠 Memory edits save as one batch — a set of memory changes is now committed as a single unit: every fact, link and rename in it lands together, including when two edits arrive at once.

🔀 Fusion answers arrive complete — a Fusion (AI analysis) answer now comes with its action buttons (copy, share, retry) and model footer straight away, and it feeds the context of your follow-up questions.

⭐⭐⭐

🌐 Research answers now end with their sources — replies from Web search and Deep research finish with a numbered list of every source, each a clickable link. No more hunting for what [7] in the text actually referred to.

📤 Image-heavy chats open light — a chat now shows a compact preview of each generated picture (about 44 KB) and fetches the full file only when you open or download it. Threads full of 4K images used to weigh hundreds of megabytes; opening one is now instant, and tapping a picture shows a dimmed loading state instead of a blank screen while the original arrives.

✍️ Ask how to make a picture and you're pointed at the right place — asking the assistant "how do I create an image" now sends you to Create image, and simply asking it to draw something switches the mode for you.

💵 Top up by card without leaving the app — the card form now opens inside the top-up popup: no bounce to an external page and back, and your balance updates as soon as the payment goes through.

🪙 Crypto top-up, all in one popup — pick the blockchain first (Ethereum, BSC, TRON, Polygon, Solana, Bitcoin and more), then the coin that lives on it, each with its own logo. You get the exact amount, the address and a QR code right there; a tap copies the amount or the address, a countdown shows how long the address stays valid, and the payment status updates live until the money lands. Back on any step returns you to the amount you had already chosen.

🪙 The minimum is shown before you commit — every coin and network has its own floor, so the popup tells you the exact minimum for the pair you picked, before you ask for an address. On the cheaper networks it can be well under a dollar.

😍😍😍😍

🔐 Lock a private chat with a PIN — any chat with a friend can now be closed with a 6-digit PIN: chat menu → Lock. A locked chat shows a padlock in the list and nothing but a lock screen until the code is entered, the same way My Intro and My Rules already work. Handy for the moment you hand someone your phone.

🧠 Your Memory Tree now learns from any chat — facts you mention in ordinary conversations — family, places, work, dates that matter — are remembered too, not just your declared interests. Tell the app something about yourself anywhere and it lands in the tree; going through Intro first is no longer required.

⚡️ Memory answers got faster and cheaper, at the same quality — the app no longer sends the same memory twice, skips an extra AI call for lookups on small trees, and shapes requests so providers can cache the stable part. An interrupted reply is no longer charged twice either.

🧹 The Memory Tree keeps itself tidy — duplicate links and repeated facts stop piling up, and records you rename or merge update in semantic search immediately. A smaller tree means sharper memory-backed answers.

🧠 The Digital Brain organizes itself — new memories pick their own hemisphere, and the map arranges them by meaning, so related memories gather into clusters instead of scattering. Long conversations also get a rolling summary, so the model keeps the thread of a long talk.

🤖 Kimi joins the main model menu — Kimi used to live only in Programming mode; now it's in the regular model picker and follows your Quality setting: Maximum quality runs the Kimi 3 flagship, Speed and Economy switch to the faster, cheaper Kimi K2.7. Worth knowing before you switch: Kimi 3 is priced like Sonnet or GPT-5.4, and K2.7 costs three to four times less — for everyday work Speed or Economy is the better value.

⏳ Incognito waits for the whole generation — a long job, like a video from a photo, no longer gets cut off. The chat waits as long as it needs and shows you the finished result, exactly like a normal chat.

💬 A generation that produced nothing tells you why — if the model couldn't make anything from your request (for example, a plain question typed while in image mode), the message now explains what happened and asks you to describe the result you want.

✅ Find anything you ever wrote — a new magnifier next to New chat opens search across all your conversations: a couple of characters in, and matching chats appear with a highlighted snippet and date. Four tabs — everything, just chat titles, just message texts, or chats with generated images. Click a result and you land on that exact message, not just the chat. Friend chats and Incognito stay private and out of search, as always.

✅ Voice chat stays with you — Audiochat no longer drops out mid-conversation. Two separate cut-offs are gone: a session used to end at around five minutes, and a few seconds of quiet while you were thinking could end it too. Pause, think, carry on — the session now closes only after five minutes of real silence, so a forgotten open mic still can't run on by itself.

✅ Reading replies aloud now costs less — the voice of any reply is about a quarter cheaper than before (~$0.04 per 1,000 characters). Exact price of every reply is in its "ⓘ" popup as always.

✅ Photos and videos in friend chats — private chats with friends now take attachments: tap the paperclip by the input, or just drop a file onto the chat window on desktop. Snaps and clips land right in the conversation, stored encrypted, and only the two of you can open them.

✅ Link previews in friend chats — share a link in a private chat and it unfurls into a card with the site's image, title and description, just like in Telegram. Works for YouTube & friends, not just light pages.

✅ Read receipts in friend chats — your messages now tick ✓ when sent and ✓✓ when your friend has read them, live. Like WhatsApp, but yours.

✅ Cleaner composer in friend chats — the AI-only "answer quality" dropdown no longer shows inside private chats with friends. Just the message field and the paperclip, as it should be.

✅ Run several generations at once — starting an image or a video no longer takes over the composer. Fire off one picture, then a video, then another angle — in the same chat, all at the same time. Each one lands in place when it's ready, wherever you've wandered meanwhile, and your balance is the only limit. One thing worth knowing: because these now run on their own, Stop no longer applies to them — an image or video you've started will run to the end and be charged for. Stop works exactly as before for replies, Fusion and Ultra.

✅ Come back to a running generation and it's right there — started something, hopped to another chat, returned? The chat shows your request still spinning away instead of looking like it vanished.

✅ Heavy chats open instantly, right where you left off — a thread full of images no longer makes you wait (and chase the bottom of the chat) while everything downloads. It opens at your latest message straight away; pictures and videos simply appear as you scroll up to them.

✅ Attach, then choose — your image goes along on the first try — attach a picture, pick an image model, hit send: the generation uses your attachment immediately, no second attempt needed.

✅ Edit your new image right away — clicking Edit on a just-generated picture (or on its retake) now instantly loads it into image editing — no more needing to reload the page first.

✅ Crypto top-ups can't get stuck anymore — the app now double-checks pending crypto payments with the provider on its own, every half hour. Even if a confirmation message goes missing on the way, your money still lands on your balance.

✅ Prices, double-checked without the noise — our background price-verifier now reads provider pages more carefully, so the rates you see in Pricing stay exactly in step with what the providers actually publish.

✅ Two models just got a lot cheaper — GPT-5.6 Luna now costs five times less than it did, and GPT-5.6 Terra 20% less. Both new rates are already in the Pricing panel and every new request is billed at them. Nothing to switch on: if Luna or Terra is your pick, your balance simply lasts much longer now.

✅ Refunds now mirror in your balance — if a card top-up is refunded or reversed by the bank, the credited amount leaves your balance too (it can briefly show negative until your next top-up). What you see is always exactly what you have.

✅ The app stays snappy while your voice replies are prepared — audio is now processed fully in the background, so listening to a long reply never slows down the rest of the app for anyone.

✅ A polite heads-up before a heavy start — starting several big generations at once (video, Fusion, Ultra)? If the balance can't cover all of them, the extra ones say so right away instead of stopping midway — no surprises, no unfinished jobs.

✅ The Gallery speaks your language now — it finally picks up the interface language you chose in Settings, just like the rest of the app.

✅ A quieter, sturdier sign-in — one fresh login link per minute per inbox (no more floods), PIN-locked panels pause after a few wrong tries, and the little things like shared-chat pages and cost details got an extra layer of polish.

✅ Chat import got sturdier — files from any source are now handled with extra care during import, so bringing your old conversations in is smooth and worry-free.

✅ A declined image now explains itself — when a safety filter turns down an image or a video, the reply says so in your language and tells you the part you can act on: whether rewording the prompt or swapping the picture will help, or whether that request will always come back the same way. Ask for something that refers to a file you didn't attach, and it says exactly that instead of failing vaguely. A generation that never produced anything is not charged — that has always been true, and now the message makes it plain instead of leaving you to wonder.

✅ Profile backups restore again — restoring a project profile (.mf) brings your Memory Tree and your saved rules back into your account. Archives written by the current app version now open as they should — a version-tag mismatch used to reject them — and older archives are still accepted.

✅ Voice sessions now have a one-minute minimum — Audiochat is billed by the time you actually speak, exactly as before. What's new: a session counts as at least one minute (roughly six cents), and you need at least that much on your balance to start one. Talk as usual and nothing changes for you.

✅ Signing in brings you back where you were — start a login from any page — Pricing, a shared chat, an invitation link — and you land back on that exact page once you're in, always inside MF0. Works the same for Google, GitHub, Discord and magic-link sign-ins.

✅ Crypto top-ups credit exactly what arrives — send part of a crypto invoice and your balance gets exactly that part, straight away. The remainder is added by itself as the rest of the payment confirms, so a top-up you finish in two goes ends up whole without you writing to support.