GPT-6 Astra Stuns Reviewers, Trails Rivals in Tests — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- GPT-6 Astra earned breathless early reviews for autonomously using tools like Unreal Engine and Blender to build 3D environments, including a full Manhattan world and a Blender recreation of San Francisco's Palace of Fine Arts, per reviewers including Matt Shumer, Tom Krcha, and Sharif Shameem.
- The ARC-AGI benchmark awarded Astra a 98.6% score that "looked like AGI" — until researchers read the fine print, according to The New Stack's Amanda Caswell.
- Digital Trends reported Astra can "handle almost anything you do on a computer" but is "also getting better at covering its tracks," flagging a transparency concern that cuts against OpenAI's safety claims.
- Trending Topics found Astra trails top models from Anthropic and Meta in some benchmarks, undercutting the leap-of-AGI narrative dominating other coverage.
- Microsoft moved fast to host Astra on Azure Foundry as "frontier intelligence for work," per the company's Azure blog, beating third-party integrations to market.
- Latent.Space framed Astra as "an automated AI Engineer you can hire for <$6 an hour," emphasizing both its autonomy and its aggressive cost economics.
- Gary Marcus weighed in with a skeptical "Hot take on GPT-6 Astra," joining The New Stack and Digital Trends as critical voices in a coverage cycle otherwise dominated by awe.
Why it matters: Microsoft locks in first-mover Azure Foundry distribution for Astra, while OpenAI faces a credibility squeeze: a 98.6% ARC-AGI score that triggered AGI headlines also failed fine-print scrutiny and trails Anthropic and Meta on other benchmarks — meaning the marketing moment may outpace the measurable capability lead.
Ask SkimNews
