Pine launches a cloud computer for AI Agent; benchmark performance and complete task completion rate diverge

AT A GLANCE

Pine Computer claims to improve AI Agent execution efficiency, but its complete task completion rate is lower than those of Codex and Claude Code.

Article

Pine gives AI built a dedicated computer: first in benchmark scores, yet behind Codex and Claude Code in actual work

Beating AI news flash, Pine AI launched AI Agent-designed cloud computer Pine Computer. It integrates cloud virtual machines,AI models, and operation tools. Developers issue tasks through API, allowing AI to operate browsers, software, and files in the cloud without using the user's own computer.

Pine's key improvement is having the computer proactively tell AI what has changed. Traditional vision-based Agents often need to repeatedly take screenshots to determine the result after a click. Pine instead lets the browser and operating system directly provide feedback on changes to the AI model executing the task, reducing the number of repeated screenshots and page checks.

Pine claims that, in internal tests, the new approach executed tasks at 2 to 5 times the speed of traditional approaches. In the SaaS-Bench v1.1 test, Pine's checkpoint score reached 78.3%, exceeding Claude Code's 74.3%and Codex's 71.1%. This score measures how many checkpoint requirements were completed during task execution, rather than the success rate of the task as a whole.

However, a note from the X community pointed out that Pine's complete task completion rate was only 27.4%, lower than Claude Code's 31.1%and Codex's 29.2%. Pine's model costs are indeed lower, averaging approximately 1.02 USD per task, compared with 20.50 USD for Codex and 26.50 USD for Claude Code. However, these costs do not include cloud server costs, and the models and execution budgets used by each approach also differ.

Pine Computer is currently in an invitation-only closed beta.

Original article link https://m.theblockbeats.info/flash/371161

Key points

Pine AI launches a cloud computer designed for AI Agent Pine Computer, which is currently in an invitation-only closed beta.
Pine Computer integrates cloud virtual machines,AI models, and operation tools, and can use API to let AI operate browsers, software, and files in the cloud.
Pine says its internal tests showed that the new approach executed tasks at 2 to 5 times the speed of traditional approaches.
In the SaaS-Bench v1.1, Pine's checkpoint score was 78.3%, higher than Claude Code's 74.3%and Codex's 71.1%。
A note from the X community stated that Pine's complete task completion rate was 27.4%, lower than Claude Code's 31.1%and Codex's 29.2%。
The report stated that Pine's average model cost per task was approximately 1.02 USD, compared with 20.50 USD for Codex and Claude Code, respectively. 26.50 USD。
The above costs do not include cloud server costs, and the models and execution budgets used by each approach differ.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

Pine Computer lets AI use a computer in the cloud and allows the browser and operating system to provide direct feedback on changes, reducing repeated screenshots and checks. In testing, it had a higher checkpoint score and lower model costs, but a lower complete task completion rate than Codex and Claude Code.

Why it matters to readers

This indicates that checkpoint scores, complete task completion rates, and model costs may reflect different aspects, so actual performance cannot be judged based on a single benchmark score.

If you want to use AI to automatically operate browsers, software, or files, you should also consider whether tasks are completed fully, whether costs include cloud server costs, and whether the testing conditions are consistent.

Risks and unknowns

The test setup and sample size for Pine's internal tests were not specified in the evidence.
The specific source of the X community note was not provided in the evidence.
The models and execution budgets used by each approach differ, so the cost comparison cannot be directly considered a comparison under identical conditions.
Pine Computer remains in an invitation-only closed beta, and its public availability and subsequent performance are unknown.

Pine launches a cloud computer for AI Agent; benchmark performance and complete task completion rate diverge

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center