GPT-6.2 Sol and GPT-6.2 Astra join the model picker, and Babbily can find an image from your chat online. Babbily can also use your connected Plugins whenever a request needs one, you decide what each Plugin tool may do, and approval cards show what they'll send and stay in the chat after you decide. You can also bring an earlier chat into your message by typing @.
GPT-6.2 Sol and GPT-6.2 Astra
GPT-6.2 Sol and GPT-6.2 Astra take GPT-6.1 Sol's and GPT-6 Astra's places in the picker. GPT-6.2 Sol is OpenAI's reasoning model for text, images and tool use, and GPT-6.2 Astra is its frontier reasoning model for coding and professional work.
Reasoning through High. Choose Low, Medium or High, or turn reasoning fully off. Neither model offers X-High or Max.
Fast mode. Select the lightning control near the composer for quicker replies. Fast mode uses more of your API usage budget.
Files. GPT-6.2 Astra reads images and PDFs directly. GPT-6.2 Sol reads images directly, and Babbily reads PDFs for it. Both have a context window of about 1 million tokens.
GPT-6.2 Astra costs less. It uses less of your API usage budget per token than GPT-6 Astra. GPT-6.2 Sol uses the same as GPT-6.1 Sol.
GPT-6.1 Sol and GPT-6 Astra are now Sunsetting. They stay selectable for now, last in the picker, with their Fast mode and reasoning controls, and chats and Scheduled Tasks that use them keep working. Choose GPT-6.2 Sol or GPT-6.2 Astra for new work.
Auto uses GPT-6.2
We tested GPT-6.2 Sol and GPT-6.2 Astra in Auto against the models they would replace, and they now handle more of your work.
Everyday conversation is quicker. Ordinary conversation now goes to GPT-6.2 Sol at a lighter reasoning level, which answered better in testing and replied faster.
Writing, research, math, planning, coding and security reviews of moderate difficulty now go to GPT-6.2 Sol, which answered about as well or better in testing and uses less of your API usage budget.
Science, finance and legal questions that aren't the hardest kind now start on GPT-6.2 Astra, which answered them as well or better in testing and uses much less of your API usage budget than Claude Fable 5.2 did.
Hard planning, debugging, server and configuration work, and data analysis now go to GPT-6.2 Astra, which answered them better in testing and uses less of your API usage budget.
Long summaries and complex database work now go to GPT-6.2 Sol, which uses less of your API usage budget. Long summaries can take a few seconds longer to start.
Hard science, finance, code generation and front-end work stay on the models that answered them best.
Everything else that used GPT-6.1 Sol or GPT-6 Astra, such as translations, SQL queries, data extraction, everyday questions and summaries, now uses GPT-6.2.
Nothing changes in how you use Auto. Learn how Auto works.
Reverse image search
Attach an image, or use one from earlier in the chat, and ask Babbily to find it online. Reverse image search looks for the image itself with Google Lens.
The same image first, then look-alikes. The reply lists the pages that show this same image, then images that only look similar, as sources with thumbnails. Say what you're after, such as the price or which bridge it is, to narrow the look-alikes.
In Auto and Manual. Ask in Auto tool mode, or choose Manual > Reverse image search. It runs only when the chat has an image, with up to three searches a reply.
Says when it couldn't search. If a search can't run or doesn't finish, Babbily tells you so instead of saying nothing was found.
Not for identifying people. Babbily doesn't use it to find out who someone is, and says it doesn't identify people from their appearance.
Some links go through Google. A few sources open through a Google link that takes you to the page.
Plugins in every chat
Babbily reaches for a Plugin when it needs one. In Auto tool mode, ask about your inbox, your calendar or a document, and Babbily uses the right connected Plugin by itself. The reply's activity shows Looking up a Plugin and Using a Plugin.
Each chat has its own switches. Every connected Plugin is on in every chat. To keep one out of a chat, select + in the composer, choose Plugins and switch it off.
@uses a Plugin right away. Type@and pick a Plugin to use it for your message. In Manual tool mode, that's the only way Babbily uses one.Shared places stay private. In public chats, team workspaces and shared projects, Babbily uses a Plugin only when you pick it with
@. With tools Off, no Plugin is used and@lists only your earlier chats.
A page for each Plugin, with Always allow
In Settings > Plugins, select Manage next to a connected Plugin to open its own page.
Every tool, in two groups. Read-only tools and Write and delete tools, with Search tools for Plugins that have many.
Choose what each tool may do. Always allow runs it without asking, Needs approval shows you a card first, and Blocked means Babbily never uses it. Change a whole group at once, or choose Reset to defaults, which asks first when a reset would unblock tools or let reads run without asking. Each change saves as you make it.
Always allow from the card. Approval cards now offer Allow once and Always allow. For a change, Babbily still asks in a few cases, such as when the Plugin changed the tool or you asked during a voice call, and the card says why. An action that ran without asking shows Ran without asking (Always allow). with Manage. Disconnecting or reconnecting a Plugin removes its Always allow choices, while Blocked and Needs approval stay.
Disconnect lives here now. After you disconnect, Reconnect becomes available on its own, without a refresh.
Know when to reconnect. If the app no longer has the account you connected, the Plugin says "Babbily can no longer reach this account. Reconnect it to use it again." and offers Reconnect.
Approval cards that show what they'll send
See the details before you allow. What Babbily will send to the Plugin lists the action's details, filled-in ones first. A required detail left empty shows as Empty, and Show everything opens every detail in full. Hidden characters, look-alike addresses and anything that looks like a password are pointed out.
Cards stay and say what happened. After you decide, a card stays where it was and says whether the action is running, finished, was cancelled, expired or was blocked. Expired cards now stay in the chat, marked as expired. Select What was sent to see what an action carried.
In order, in your language. A reply's cards keep the order its steps asked for them, and each card asks its question in your language.
Requests with several steps carry on. After you approve one step of a request such as "create a label, file the email under it and reply", Babbily continues to the next step in the same reply.
Emails open right away. Allowing a Gmail card that sends an email shows the email for review without a reload. If it was changed in another tab, Babbily shows it again before sending. Esc cancels an email only when you press it inside the email, and when something fails, Babbily says what couldn't be done, such as Couldn't send this email or Couldn't save your changes.
Your email stays yours. People who can see a shared or team chat see where an email stands, but not its recipients, subject or body.
More Plugin tools work. Some tools Babbily couldn't find before now work from chat, such as Gmail's Move to Trash and many GitHub, Excel, Slack and Outlook tools. Large results now come through, and asking twice for the same action with different details gives each its own card.
Bring an earlier chat into your message
Type @ in the composer and pick one of your earlier chats, or paste a link to one. The reply reads that chat, a summary of its earlier part and its latest messages, so you can pick up where you left off without pasting anything in.
Your own chats, from the same place. Pick any of your chats from the same workspace, and from the same project when you're in one, up to three in a message.
Memory on or off. It works either way, and Babbily reads exactly the chats you mention.
Not in shared places. Temporary Chat, public chats, chats shared with your team and team projects don't read other chats. If a chat can't be read, the reply says which one and why.
Rate a reply
A thumbs down now asks Tell us why. Choose Not accurate, Didn’t follow my request, Incomplete, Harmful or offensive, Wouldn’t help, Something broke or Other, and add an optional note. What you send goes to the Babbily team, who read it to improve Babbily. Rate a reply.
Voice Mode uses your Plugins
Plugins during a call. A call can use your connected Plugins by itself, as a chat in Auto tool mode does. Anything that needs your approval shows a card on screen.
Calls stay connected through a card. The microphone turns off while a card waits, and the call no longer drops. It comes back on after you answer: once the action has run after Allow once, or a moment after Cancel. If you muted yourself, a card doesn't unmute you.
Ending a call cancels its waiting cards. End a call while a card waits and the card shows Cancelled. Nothing was sent to Gmail. (with your Plugin's name) right away, with no tap. An email waiting for Send shows Email not sent. You can type, or start a new call, straight away.
Muting right after you ask keeps the call. If you mute just after asking something, the call stays connected and Babbily still answers. Finish your sentence before you mute: a sentence cut off by the mute isn't sent.
A clearer message in a noisy room. If Voice hears talking for about 30 seconds but can't make out a request, it says Voice couldn’t make out what you said. It may be noisy where you are. Tap Continue to try again. instead of saying the connection dropped.
Voice Mode is available on paid plans. Use your Plugins during a call, approve requests during a call and talk naturally.
Auto judges files, projects and skills too
A message with a file, a chat in a project, and a message that uses a skill used to go to Auto's dependable default. Auto now judges how demanding the request is from what you wrote and picks a model at that level, so a hard question can still get a stronger model. Later messages in a chat with a file are judged like any other. Learn how Auto works.
Also in this release
No stray Web search step. Replies that use Places or Reverse image search no longer show a Web search step in their activity when no web search ran.
Approve shared memory. In Settings > Team, an older shared memory marked Needs review can now be approved by its author or the team owner. Manage memory.
Memory loads right after you sign in. The Memory tab in Settings now loads as soon as you're signed in.
Scheduled Tasks pause instead of getting stuck. A task whose Plugin connection lapsed now pauses with Plugin reselection required. Reconnect the app, save the task, then resume it, since saving doesn't resume it. Reconnect Plugins for a task.
Images that never finish now end. An image that's still not done ends with Image generation took too long. Please try again. within about 15 minutes, so you can try again.
Cut-off artifacts say so. An artifact that stopped partway now says This artifact took too long to generate. above what was written, instead of looking finished.
Stopped artifacts appear right away. Stopping a reply while an artifact is being written shows its card at once, without a reload.
Send to Google Docs works the first time. Sending an artifact to Google Docs from the artifact panel or the Artifacts page works even if you haven't used Google Docs in chat first.
Translate no longer gets stuck. Text you type while Translate is still signing you in is translated once you're signed in, instead of staying on Translating....
Email cards say when your choice came too late. If an email was already sent or cancelled in another tab or on another device, choosing Send, Cancel or Edit on an older copy now says so, for example This email was already sent, and the card shows where the email stands. Approve actions before they run.
