✅ Chat Navigator — in long conversations a new button now appears at the bottom right of the chat. One tap opens a scrollable table of contents: all your questions in order, each trimmed to a short preview. Tap any entry to jump straight to that message. Appears automatically once there are more than 4 questions in the dialog.

✅ Under the hood — faster chat navigation (switching dialogs now cancels the previous background load immediately), plugin jobs survive server restarts without data loss, and a round of async and query-plan improvements for cleaner performance under load.

✅ A list of minor bugs has been fixed, but optimizations have also been implemented that will allow you to save up to 3% of your money. The project's goal, as always, is to ensure that users pay as little as possible.

✅ Replies can now match your Digital Brain — there's a new switch in Settings, "Digital Brain influence on replies" (off by default). Turn it on and the assistant adapts its tone to how your memory is balanced: if your Digital Brain leans logical and analytical, answers become more concise and structured; if it leans personal and creative, they become warmer and more conversational. It's a gentle stylistic nudge only — it never changes the facts or overrides your instructions. Kicks in once your Digital Brain has enough classified memory.

✅ A clear heads-up when a chat gets too long — very long conversations can pass the model's maximum size (200,000 tokens), and the request would otherwise fail with a cryptic error. Now you'll see a simple popup explaining that the current thread plus your prompt are over the limit, and suggesting you start a new chat to continue. Tap OK (or click outside) to close it.

✅ AI opinion uses all your models again — when you ask several models at once, Seed now takes part in the panel alongside the others. We'd briefly paused it while tracking down an unrelated issue; it's fully back.

✅ Shared messages now carry their image into link previews — when you share a message or a conversation that contains a generated image, that image now shows up as the preview thumbnail on social networks and messengers (Telegram, X, and others), instead of a generic logo. Share a message with a video and we automatically grab a still frame from it to use as the preview picture. There's nothing to set up — it just works the next time you share, so your links look like what you actually made.

https://www.mf0.ai/share?msg=c8f65102-46dc-4ccb-9fc9-0b7c26b6c14b&rep=1&ref=5FD4CD10A7

🚀 A new branch of MF0.ai is here — Programming. This is the first step of a whole new direction for the project: tools built for writing and shipping code. More is on the way — here's what's live today.

✅ Programming mode — open the "+" menu and choose Programming. The composer switches into a focused coding mode (the quick-action banners step aside), and you pick any model from the dropdown to start building.

✅ New code-focused models — alongside Gemini, ChatGPT, Claude and Grok, Programming mode adds three models tuned for code:
• Nemotron 3 Ultra — a large, high-capability model for complex tasks
• Nemotron 3 Nano Omni — a fast reasoning model
• Kimi K2.7 — a coding-specialist reasoning model

Nano Omni and Kimi "think" for a moment before answering, so the first words may appear after a short pause — that's normal. And answers can run long: the output limit is raised in this mode, so you get complete files, not truncated snippets.

✅ Download code as files — every code block in a Programming reply gets a download bar. Save a snippet straight to disk with the correct extension (.py, .js, .ts, .html, …) and a sensible filename. Code is syntax-highlighted — and the same works on shared pages: share a reply and the recipient gets highlighted, downloadable code too.

This is just the beginning of MF0.ai's programming branch — more developer tools are coming. 🛠

❇️ An example screenshot can be seen here

✅ More money-saving optimizations — we've tuned how generations are processed under the hood, so many requests now cost even less. The project's goal, as always, is for you to pay as little as possible.

✅ See the exact cost of any reply — tap the "i" on a message to open a full breakdown: which model answered, the tokens used, and the precise price down to the cent. It now covers everything — text, image and video generations, and Incognito too.

✅ Crystal-clear billing — your cost breakdown, your balance, and your analytics are all driven by the same verified pricing, so the numbers always line up wherever you look.

✅ Friendlier Incognito — features that need an active conversation (like Fusion and video-to-video) now greet you with a simple heads-up instead of a technical error.

✅ AI opinion is more focused now — by popular request, we've removed Seed from the AI opinion panel. Several of you found it pulled the discussion off-track and got in the way of the other models. When you ask the panel now, you get cleaner, more on-point answers from the rest of your models. Seed remains fully available as usual in regular chat — this change only affects the AI opinion set.

✅ One-tap AI modes — Settings → AI settings now has four presets: Maximum savings, Maximum speed, Maximum quality, and Custom. A single tap sets the right model for every task — dialogue, images, search, and research — across every provider. Pick savings, speed, or quality and you're done. Adjust any single model by hand and it switches to Custom; press Save AI models to keep your choice.

✅ "Image generation analysis" switcher — a new toggle (off by default) that lets your Memory Tree also learn from your image requests. Useful if your main workflow is visual content; leave it off for everyday use to keep image requests cheaper.

✅ Plain-language hints everywhere — every setting now has a short explanation right underneath, so it's clear what each option does before you touch it. Clearer names too: "Interface scale" → Font size, "Chat analysis" → Light model, "AI priority" → Priority model, "Digital Brain influence" → Digital Brain.

✅ Cleaner Settings screen — bigger section headings and a tidier layout make everything easier to scan.

✅ AI Opinion is more resilient — when you ask several models for their take at once, a single model failing (too busy, or a very long prompt it can't accept) used to break the whole panel and you'd get nothing back. Now a model that can't answer is quietly skipped and the rest of the opinions still come through — you always get the answers that worked.

✅ Fusion gives one clean answer again — Fusion blends several models' research into a single, unified write-up. A glitch had it falling back to stitching the raw answers together, so you'd see the same points repeated three or four times in slightly different words. Fixed — Fusion now returns one coherent, de-duplicated document, the way it's meant to. And if the blending step ever hiccups, it now says so clearly instead of quietly padding the reply.

✅ Security hardened — your conversations are now sealed to your account. The text of a model's reply, your prompts, your generated images/videos and your voice transcriptions can be read by no one but you — there's no path for another account to intercept them.

✅ Sharing shows only what you share — a public link now exposes strictly the chat or single reply you chose to make public; private conversations never surface, even to someone probing links directly.

✅ Plus 7 more hardening and reliability fixes shipped behind the scenes this round.

✅ More reliable replies — fixed an edge case where unusual saved-memory data could make a message come back empty. Responses now go through reliably.

✅ Your custom rules, applied faithfully — saved conduct rules are reliably kept even when edited in quick succession, and the assistant correctly honors your highest-priority rules when composing each reply.

✅ Lighter Memory Tree — repeatedly opening and closing the 3D Memory Tree no longer leaks memory, so long sessions stay smooth.

✅ Fusion: faster, leaner, sturdier — Fusion (multi-model deep research) now verifies all sources in one web-grounded pass instead of one check per source, so a run is quicker and costs less for the same grounded quality. It also automatically routes around a temporarily-unavailable model, so a run never gets stuck.

✅ Fully localized interface — every part of the UI now follows your language, down to the small details: button tooltips, icon labels, the gallery, the shared-conversation page, voice chat, and the analytics screen. All 11 languages are covered with no leftover English text.

✅ AI opinion now in Help — you can now get a multi-model AI opinion right inside the Help assistant: pick AI opinion in the model selector and every connected model answers your question in one panel, capped with a short combined summary. Handy for a well-rounded take on how to use the app.

✅ Smoother panel switching — opening Help or the Memory Tree straight from Settings now switches cleanly, instead of leaving the Settings screen lingering behind.

✅ Over 20 minor technical fixes have been implemented, including service performance optimizations.

✅ GPT-5.5 Pro now available — the GPT-5.5 Pro reasoning model now works in chat (and live streaming). Pick it in the model selector — or the "quality" preset — when you want the most thorough, deeply-reasoned answer.

✅ No accidental double-sends — sending a message is now guarded so a quick second click or Ctrl+Enter while it's starting can't fire a duplicate request.

✅ Steadier long chats — fixed a rare case where trimming a very large conversation to fit the model could make the page hang. Long chats now stay responsive.

✅ Clear feedback on video/audio files — if you attach a video or audio file in a normal chat, the app now tells you upfront that it won't be read by the chat model, instead of quietly leaving it out.

✅ Up-to-date model lineup — refreshed the model list: Grok now runs the latest Grok 4 model, and a few retired models were removed. If you had one of the old ones selected, it automatically switches to its current replacement — so you're always on a working, up-to-date model.

✅ No more silently cut-off code — long answers (like a big HTML/code file) now run much further before hitting any length limit, so they rarely get cut. And if one ever does, you get a clear notice that the reply was cut short with a one-click Continue button right under it: tap it and the assistant resumes exactly where it stopped — no repeating, no retyping your request. You only pay for what's actually generated.

✅ AI opinion no longer loses your answer if you stop early — if you interrupt a multi-model AI-opinion panel mid-way (switch chats, reload the page), the answers collected so far are now saved instead of disappearing. You keep the partial result rather than an empty chat.

✅ Steadier AI analysis (Fusion) — the final combined answer is more reliable: if the summarising step is momentarily unavailable, Fusion automatically falls back to another model and still returns a proper unified write-up instead of a degraded raw dump.

✅ Interface polish — the Share button now uses the standard share symbol; copying a share link shows a clear "Link copied to clipboard" message; and the per-chat actions menu (Share, Rename, Delete…) is now visible right away on every chat in the sidebar instead of only appearing on hover.

✅ Reserve your nickname — Settings now has a Nickname field where you can claim your own unique @handle (4+ characters, Latin letters, digits and underscore). It checks availability as you type, and Apply only lights up when the name is free. This is the groundwork for interacting with other users inside the project — chatting, sharing AI models together, and tagging friends in AI chats.

✅ Claude Sonnet 5 is now available — the newest Claude Sonnet model can be selected for chat, web search and deep research. It's Anthropic's latest Sonnet — a strong all-round pick — and it's fully wired into billing, so the per-message cost shown in the "i" popup is exact.

✅ Reuse any past request in one click — every message you sent now has a round-arrow button (↻). Tap it and the composer is refilled with that exact request — prompt, mode, model, attachments and all image/video settings (aspect ratio, resolution, quality, duration) — ready for you to tweak and resend. Nothing is sent automatically and the original message stays untouched. Works even after reloading the chat.

✅ Pick image resolution & quality — when creating images you can now choose the output resolution (1K / 2K / 4K) for Nano Banana models and the quality (Low / Medium / High) for the ChatGPT image model. Only the options that apply to your selected model are shown, so there's no clutter.

✅ Choose the image aspect ratio — in Create image you can now pick the shape (16:9, 9:16, 1:1, 4:3, 3:4) for ChatGPT and Gemini, and it's actually applied to the result. The picker only appears for models that support it, and the ↻ "reuse request" button restores your choice.

✅ Image & video generations survive a server update — if the app restarts while your image or video is still being generated, the job is now picked up and finished automatically instead of getting stuck. Your result comes through, and you're only charged once it's actually delivered.

✅ Stop now stops instantly in AI opinion — pressing Stop during a multi-model AI-opinion panel halts every model at once, right away — no more waiting for the current model to finish first.

✅ Incognito fixes — the incognito transcript now correctly remembers which model produced each answer, the cost (i) button works on every result, and you can regenerate an incognito answer (↻). Favourites and Share stay disabled in incognito on purpose — the session is private.

✅ Easier model selection on mobile — in Settings, the per-family AI model pickers now stack cleanly on phones: each provider (ChatGPT, Claude, Gemini, Perplexity, Grok) shows as a heading with its model dropdowns full-width right below, so you can read the model names in full instead of seeing them cut off.

✅ Your @nickname in the profile menu — once you've reserved a nickname (Settings → Nickname), the profile popup now shows your @nickname at the top instead of your email — a friendlier identity that also keeps your email more private.

✅ New "friends" achievements — the Achievements panel gains a set of social medals: earn one just for reserving your nickname, plus more for your first mutually-confirmed friend and for reaching 10 / 50 / 100 friends. If you already picked a nickname, the nickname medal is unlocked for you automatically.

✅ Add friends by @nickname — a new Friends list (Profile → Friends list) lets you find people by their @nickname and send a friend request. The other person confirms it from their side, so friendships are always mutual. Handy status filters let you see incoming requests and confirmed friends at a glance.

🔥 Private chats with friends — once you and someone are friends, the "+" menu gets a Chats with friends option: message each other privately, right inside MF0. Friend chats sit in their own folder, arrive in real time, and your messages are stored encrypted. Links you send are clickable and show a rich preview card when available, and deleting a chat clears it for both of you.

✅ Refreshed home screen — the new-chat home now greets you with redesigned quick-action cards for Create image, Fusion (AI analysis), Programming, Deep research and Web search — each with its label above a crisp new illustration, with matching light- and dark-theme artwork.

✅ Reach Help & the Memory Tree from anywhere — opening the Help assistant or the 3D Memory Tree now works cleanly from the Gallery, Pricing and Achievements screens too, instead of leaving them lingering on top.

✅ Simpler image-model list — when creating images, every image model now appears in one flat list (including plugin models), with no confusing separator — pick any of them in a single place.

✅ New "News" section — the profile menu now has a News page (right above Help) that gathers all of MF0.ai's updates in one place: newest first, in your interface language, with older entries loading as you scroll. Same updates as our channel — now right inside the app.

✅ Two new Image → Video models — the video generator gains Happy Horse and Gemini Omni Flash, joining Kling 3, Seedance 2, Veo 3.1 and Wan 2.7. Pick either one in the model selector when animating a photo; per-second pricing is listed in Pricing and the exact cost of each clip shows in the "i" popup.

✅ "Reuse this request" sends in one click — hitting the round-arrow (↻) on a past request now fills the composer and enables Send immediately, so you can resend right away — no need to type an extra character first to "wake up" the button.

✅ No duplicated attachments when reusing a request — clicking ↻ "Reuse this request" several times no longer stacks up copies of the attached images; each reuse cleanly replaces the attachments with the original set.

✅ Composer prompt hint resets correctly — after picking a quick-action like Create image, starting a New chat now restores the normal "Type your message…" hint instead of leaving the mode-specific one (e.g. "Describe the image…") stuck in the input.