A month ago we let octos borrow your login: an agent hit a sign-in page, you signed in yourself, and the agent carried on in the browser you left behind. That browser was Chrome running inside your machine, competing with your octo's work for the machine's memory.
Every octo now has its own cloud browser. It's a remote browser that runs on our side, not on your machine. You choose where in the world it browses from, you can watch it work or take the keyboard, and it asks you to sign in only when a site actually shows a login, a 2FA prompt or a CAPTCHA. This post covers everything new in Octomind Cloud since our October 25 newsletter.
Happy Halloween: $20 of credits, on us. Use OCTO-HW-V7CHDT on any plan, Free included, once per account, for the first 250 people, until December 31. Spend it on a job you've been putting off, then tell us how it went.
A cloud browser for each octo
Ask an octo to check a supplier portal, compare prices across three shops or pull a report from a dashboard with no export, and it opens a browser. Here's what that AI agent browser is now:
- One per octo, and it remembers. Each octo has its own browser profile, so the sites you signed in to stay signed in the next time it opens one. Two octos don't share cookies.
- A real browser, not a scraper. Pages render as they do for you, scripts and all, so your octo's web browsing sees what you would.
- It runs on our side. Chrome no longer runs inside your machine, so browsing doesn't take memory from the work your octo is doing there.
- It closes on its own when the octo is done with it.
Pick where it browses from
Under your octo's Setup tab, Browser location offers:
| Option | What sites see |
|---|---|
| United States | The same US address every time the octo opens a browser |
| United Kingdom | The same UK address every time |
| Germany | The same German address every time |
| France | The same French address every time |
| Singapore | The same Singapore address every time |
| No fixed address | A new address each time |
A fixed address helps on sites that challenge logins from a new place, and on shops and search results that change by country. A change applies the next time the octo opens a browser.
Watch it work, or take over
While an octo is browsing, its Work tab says so, with a Watch or take over link. It opens the live browser in the panel: watch the octo click through, or take the mouse and keyboard yourself, then hand it back.
Sign in only when a site asks
Octos used to ask for a sign-in up front. Now they open the site first, and ask you only when a login page, a 2FA prompt or a CAPTCHA actually appears. The Work tab then shows A browser is waiting for you to sign in to {site}. Open it, sign in, and the octo carries on from where it stopped.
The boundary from the original browser-login post still holds: your password goes from your keyboard to the site and never reaches the model, but the octo does get the signed-in session. Give it accounts you'd give a contractor.
What it costs
The cloud browser is part of the paid plans:
- $0.02 each time a browser starts.
- $0.015 per minute it's open, with the last minute rounded up when it closes.
A ten-minute errand on one site costs about 17 cents. On Free, octos don't get the browser; signing in to sites needs a subscription. Prices for everything are on the pricing page.
See what a conversation cost
The conversation header now shows Today $X · Total $Y. Click it for Conversation spending: that conversation's model costs, day by day, newest first. Days are in UTC. Machine time and activation fees aren't included, so the number is exactly what the conversation's model calls cost you.
It's the quick answer to "why did my balance move today?": open the conversation you were in and look.
Octos ask before they guess
Some tools need something only you know: which environment to deploy to, what size to order, whether to go ahead with a refund. Your octo now asks in a form, Your octo needs a few details, with the name of the tool that's asking. Fill it in and Submit, or choose Don't share or Not now. A request to open a link shows the site's address first and warns you to open it only if you trust it.
On the other side, octos stop asking about the obvious. A follow-up that is clearly part of the job you gave gets done without a "shall I?", with at most one suggested next step at the end. And routines that run on a schedule now report what they did, instead of ending on a question nobody is there to answer.
Talk, quote, and command
- Press-to-talk. Hold the mic button to record while you hold it, or tap to start and tap to stop. The recording is saved in your browser before it uploads, so a failed upload, a reload or leaving the chat doesn't lose it. Retry on the attachment sends it again.
- Quote a reply. Select text in a reply, or hover a block, and click Reply. The quote lands in your message, so the octo knows exactly which line you mean.
- Slash commands. Type / to see your octo's commands, filtered as you type.
- Stop sits beside the box. A reply you stop ends with a quiet Stopped. and a Send again button, not an error.
- See what it's thinking. While an octo works, the tail of its reasoning shows under the working label, so a long pause has something to read.
- Long chats open faster. Older messages load when you click Show earlier messages.
More model calls at once on paid plans
When an octo hands a job to several specialists, they call models at the same time. Paid plans now run more of those calls in parallel:
| Plan | Before | Now |
|---|---|---|
| Pro | 3 | 5 |
| Max | 5 | 10 |
| Team | 15 | 20, shared by the team |
A call over the limit waits in line for up to 20 seconds (it was 15) before it's turned away.
Models
| Model | Price per 1M tokens (in / out) | Context window | Note |
|---|---|---|---|
| GPT-6.1 Terra | $1 / $5 | 1M | Replaces GPT 5.6 Terra ($2 / $12) |
| MiMo V2.6 Flash | $0.14 / $0.28 | 1M | Xiaomi |
| MiMo V2.6 Pro | $0.435 / $0.87 | 1M | Xiaomi |
- On every plan now: Kimi K3, Kimi K3.1 Pro and Qwen3.8 Max.
- Retiring: DeepSeek V4 Flash on November 1; DeepSeek V4.1 Flash ($0.30 / $1.20) takes its place. GPT 5.6 Terra retires on November 10; GPT-6.1 Terra costs less.
- Specialists start on models picked for their job. Out of the box, each kind of specialist runs a model suited to it, for example Kimi K3.1 Pro for finance and Gemini 3.8 Flash for legal. Change any of them under Specialist models in Config.
Each model has its own page on the models page.
Images, video and voice
- New from MiniMax: Image-01 at $0.0035 per image, H3 Lite and H3 Highspeed video at $0.04 and $0.06 per second, and Speech 2.8 HD and Turbo at $0.003 and $0.0018 per second. Speech 2.8 HD is now the default voice, and you can pick which voice it speaks in.
- Price changes: Wan 3.0 goes from $0.10 to $0.20 per second, and GPT Image 2 from $0.13 to $0.15 per image, so they cover what the provider charges us. Eleven Music drops from $0.015 to $0.01 per second, billed per minute.
Smaller things
- A public profile. Under Settings → Account → Public profile, add an avatar, a username, a display name and a short bio. The sidebar shows them.
- Clearer billing history. Promo credits, gifted plans and welcome credits each get their own label.
- Queued messages wait out a billing hold. If your account is paused for usage or a payment, queued messages stay queued and go out once it's sorted.
- Machines run Octomind 0.67.0 from their next start. Model calls travel over one persistent connection, and an octo reads several files at once instead of one by one. The 0.67.0 post covers what changed inside.
Tell us what broke, what confused you, and what you expected an octo to do that it didn't. Post in Discord or reply to our newsletter. A screenshot and the time it happened is usually enough for us to find it.
— DK
