[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
 Nick Dobos [@NickADobos](/creator/twitter/NickADobos) on x 49.8K followers
Created: 2025-07-17 19:21:37 UTC
chatGPT Agent recap:
New Agent mode, has access to
- A text browser, similar to deep research & web searching
- A "computer use" GUI browser, similar to operator, visual input, clicking & dragging
- A Terminal, allowing the ability to run commands, manipulate files, access to APIs, and data connectors (MCPs under the hood)
- This also allows making files, like power points & spreadsheets
- Image generation, either via imagen models, or via code for charts similar to data analysis mode.
Blends ideas between deep research and operator. For example, Operator struggles to read long articles, it needs to keep scrolling down, but deep research can read the whole thing. Conversely Deep research struggles with interactive elements and highly visual webpages, where operator can easily see them and take action. As well as a virtual file system.
Aimed at long runnings tasks, sometimes taking 10+ minutes
+ the ability for the user to interrupt and modify as the task runs.
+ the ability for the model to ask follow up questions, to stop and ask the user sign in into things or to confirm final steps
Allows deep researching auth'ed and logged in websites, as well as jumping into the the agent's virtual browser to take manual steps.
Final output includes a scrubbable history showing the steps taken.
These tools increase benchmark scores, on answering questions, agentic browser use benchmarks, editing spreadsheets, as well as real world tasks, for example a benchmark of investment bankers tasks like putting together reports and financial models.
They mentioned X separate disclaimers about prompt injection, despite their various attempts to monitor and mitigate. Lots of warnings.
Launching today for Pro users, XXX queries a month. Rollout should be finished EOD.
Plus & teams starting today. XX queries a month. Rollout finished "very soon".
Edu & Enterprise later this month.
XXXXXX engagements
**Related Topics**
[agentic](/topic/agentic)
[hood](/topic/hood)
[files](/topic/files)
[gui](/topic/gui)
[open ai](/topic/open-ai)
[Post Link](https://x.com/NickADobos/status/1945926883943882790)
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
Nick Dobos @NickADobos on x 49.8K followers
Created: 2025-07-17 19:21:37 UTC
chatGPT Agent recap:
New Agent mode, has access to
- A text browser, similar to deep research & web searching
- A "computer use" GUI browser, similar to operator, visual input, clicking & dragging
- A Terminal, allowing the ability to run commands, manipulate files, access to APIs, and data connectors (MCPs under the hood) - This also allows making files, like power points & spreadsheets
- Image generation, either via imagen models, or via code for charts similar to data analysis mode.
Blends ideas between deep research and operator. For example, Operator struggles to read long articles, it needs to keep scrolling down, but deep research can read the whole thing. Conversely Deep research struggles with interactive elements and highly visual webpages, where operator can easily see them and take action. As well as a virtual file system.
Aimed at long runnings tasks, sometimes taking 10+ minutes
- the ability for the user to interrupt and modify as the task runs.
- the ability for the model to ask follow up questions, to stop and ask the user sign in into things or to confirm final steps
Allows deep researching auth'ed and logged in websites, as well as jumping into the the agent's virtual browser to take manual steps.
Final output includes a scrubbable history showing the steps taken.
These tools increase benchmark scores, on answering questions, agentic browser use benchmarks, editing spreadsheets, as well as real world tasks, for example a benchmark of investment bankers tasks like putting together reports and financial models.
They mentioned X separate disclaimers about prompt injection, despite their various attempts to monitor and mitigate. Lots of warnings.
Launching today for Pro users, XXX queries a month. Rollout should be finished EOD. Plus & teams starting today. XX queries a month. Rollout finished "very soon". Edu & Enterprise later this month.
XXXXXX engagements
