PulseAugur
EN
LIVE 16:11:51

LLMs tested in sandbox: Claude, Gemma, and gpt-oss build, others test

By PulseAugur Editorial · [1 sources] ·

An experiment tested eight large language models by giving them a sandbox environment with no specific task, only a set of tools and 50 actions. The models diverged into two main groups: 'builders' that created new tools or scripts, and 'testers' that explored the available tools before stopping. Claude, Gemma, and gpt-oss were categorized as builders, with Claude producing one finished tool per run, Gemma creating complex projects, and gpt-oss generating demo scripts. Qwen and Granite acted as testers, performing basic checks on each tool before concluding their runs. AI

IMPACT This experiment reveals differing intrinsic behaviors and capabilities of LLMs when given autonomy, highlighting potential differences in their underlying architectures or training objectives.

RANK_REASON The item describes an experiment testing the behavior of multiple LLMs in a controlled environment, which falls under research. [lever_c_demoted from research: ic=1 ai=1.0]

Read on dev.to — LLM tag →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

LLMs tested in sandbox: Claude, Gemma, and gpt-oss build, others test

How we ranked this

Signal score
35 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
The item describes an experiment testing the behavior of multiple LLMs in a controlled environment, which falls under research. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
model release
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. dev.to — LLM tag TIER_1 English(EN) · vadim albarov ·

    I gave 8 models a sandbox and no task. Here is what each one did with the freedom.

    <p>Every benchmark tells a model what to do. I wanted to see what happens when nobody does.</p> <p>So I gave each model its own Linux container, five tools, 50 actions, and one message: there is no task, do whatever you want, nobody will answer. Seven local models from seven comp…

RELATED ENTITIES

RELATED TOPICS