Pine AI has introduced Pine Computer, a cloud-based computer built for AI agents. The product combines cloud virtual machines, AI models and operating tools, allowing developers to send tasks through an API so an AI system can operate browsers, software and files remotely instead of using a user’s local machine.
Pine says its main improvement is giving the model direct updates from the browser and operating system, rather than relying on repeated screenshots to detect what changed after each action. According to the company, internal tests showed the setup could run tasks at 2x to 5x the speed of traditional vision-based agents.
On SaaS-Bench v1.1, Pine posted a checkpoint score of 78.3%, ahead of Claude Code at 74.3% and Codex at 71.1%. That metric measures how many intermediate checks are completed during a task, not whether the entire task is finished successfully. Notes shared on X said Pine’s full task completion rate was 27.4%, below Claude Code’s 31.1% and Codex’s 29.2%.
Pine also reported lower model costs, averaging about $1.02 per task, versus $20.50 for Codex and $26.50 for Claude Code. Those figures exclude cloud server costs, and the compared systems used different models and execution budgets. Pine Computer is currently in invite-only testing.
Pine AI has rolled out Pine Computer, a cloud computer designed for AI agents. The system combines cloud virtual machines, AI models and operating tools, letting developers send tasks through an API so the agent can operate browsers, software and files in the cloud rather than on a user’s own computer.
Pine targets a bottleneck in vision-based agents
Pine said its main change is that the computer tells the AI what has changed. Traditional vision-based agents usually need to keep taking screenshots to work out what happened after a click. Pine instead has the browser and operating system send updates directly to the model carrying out the task, cutting down on repeated screenshots and page checks.
The company said internal testing showed the new setup could execute tasks at 2x to 5x the speed of conventional approaches.
Higher checkpoint score, weaker full completion rate
In SaaS-Bench v1.1, Pine recorded a checkpoint score of 78.3%, ahead of Claude Code at 74.3% and Codex at 71.1%. That score measures how many checkpoint requirements are completed during task execution, rather than the success rate for the entire task.
But notes circulated on X said Pine’s full task completion rate was 27.4%, below Claude Code’s 31.1% and Codex’s 29.2%.
Lower model cost per task
Pine’s model cost was also lower, averaging about $1.02 per task. Codex and Claude Code came in at $20.50 and $26.50, respectively.
Those figures do not include cloud server costs, and the systems being compared used different models and execution budgets.
Pine Computer is currently in invite-only internal testing.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
