이미지: X — 모델·오픈소스 화면 갈무리
Summary
- MiniMax released the v3.0.54 update to its coding agent app "MiniMax Code" on August 11, 2026
- A new "Browser Control" feature lets the built-in browser read pages, fill out forms, and automatically verify results
- The update also adds "Goal Mode," which tracks progress on long-running tasks set as "goals," along with a Feedback Center and customizable keyboard shortcuts
- 버전
- MiniMax Code v3.0.54
- 공개 시점
- 2026년 8월 11일(UTC), X 공식 계정 발표
- 신규 기능
- 브라우저 컨트롤, 목표 모드, 피드백 센터, 파일 출력 개선, 키보드 단축키
- 화면 표기 모델
- MiniMax-M3 (스크린샷 내 'Thinking | MiniMax-M3' 표기)
- 직전 버전
- MiniMax Code 2.0, 2026년 8월 7일 공개, 오픈소스 Pi Agent 프레임워크 기반 재구축
A coding agent that fills out login forms for you
Until now, coding agents were limited to writing code and editing files. The v3.0.54 update to "MiniMax Code," released by MiniMax on August 11, pushes that boundary into the browser. Using the app's built-in browser, the agent can now read web pages directly, fill out forms such as sign-ups or checkout flows, attach files from the working folder, and upload them. Once a task is done, the agent also verifies the results itself. MiniMax described the feature on its official X account as reading the browser, filling out forms, and automatically checking results.
What's new
The company said this update bundles several features that users had frequently requested. There are five core additions.
| Feature | Description |
|---|---|
| Browser Control | Automates page reading, form filling, file uploads, and result verification in the built-in browser |
| Goal Mode | Registers complex, long-running tasks as "goals," with continuous status and elapsed-time tracking |
| Feedback Center | Report issues from the chat window or menu, with task, workspace, and app version auto-attached |
| Improved file output | One click jumps directly to the folder or code line for generated or referenced files |
| Keyboard shortcuts | Searchable, customizable, and resettable from a dedicated Settings page |
According to screenshots shared by the company, Browser Control can be toggled on or off on screen and sits alongside a separate switch for the "Computer Use" feature. Goal Mode also appears as its own item in the same menu, suggesting that browser operations and long-task management can be enabled independently.
Goal Mode and Feedback Center: built for trusting the agent over the long haul
Goal Mode is designed to let users hand off a task to the agent and stop worrying about it. Once a complex task is registered as a "goal," the agent keeps working toward completion, and users can check progress and elapsed time at any time. The biggest source of unease when delegating long autonomous tasks is not knowing what the agent is currently doing — this feature appears to be an attempt to ease that unease with a single status panel.
The Feedback Center, conversely, is built for when the agent gets something wrong. Users can report issues directly from mid-conversation or from a menu in the bottom-left corner of the screen, with the current task, workspace, and app version automatically attached. Screenshots and diagnostic logs can also be attached, and after submission, tickets can be checked under "My Feedback." Screenshots shared by the company show the report screen includes a description field (up to 1,000 characters), screenshot attachments (up to six), and an optional contact field.
Background: recent developments in MiniMax Code
MiniMax Code underwent a full rebuild as version 2.0 on August 7. At the time, the company said it had rebuilt the app on top of the open-source agent framework Pi Agent (@pidotdev), emphasizing stability improvements across three areas: everyday conversation, office tasks, and long, complex tasks. The following day, August 8, MiniMax announced that agent plugin support would be available from launch. v3.0.54 arrives less than a month after that rebuild, and rather than simply following through on the stability improvements promised in 2.0, it leans heavily toward features users directly requested, such as browser automation and task tracking.
MiniMax is not alone in giving coding agents the ability to operate browsers. Across the coding agent industry, there is a broader push to have agents handle not just code generation but tasks that require going through a web UI to complete — such as logging into a live service to check deployment status or registering an API key.
What changes as a result
With this update, users can now hand off browser-based tasks — such as signing up for and configuring web services — to the same agent that writes their code, in a single continuous workflow. For example, the entire process of signing up for a new API service, obtaining a key, and wiring it into code can now be handed off without human intervention. Goal Mode lowers the barrier to trusting the agent with hours-long refactoring or migration work by letting users check progress in real time. The Feedback Center, meanwhile, acts as a safety valve to absorb the growing burden of error reports as automation scope expands. As more coding agents gain the ability to operate browsers directly, whether this kind of status tracking and reporting infrastructure keeps pace is likely to become the real test of trust.
