GPT-6 Astra in Practice: Tests, Use Cases, and Feedback

A video puts GPT-6 Astra and Claude Opus 5.5 side by side for anyone who wants to see how the two frontier models compare in practice.

Mollick shares a book sales video that Astra made in Blender, narrated from the book's own perspective with a vaguely unsettling tone.

In a 30-task agent evaluation, Pareto tied Astra for first place, with a cost per successful task of $0.52 versus $1.58.

The author says Astra alone wrote Photon Studio, a free cross-platform image editor with 20k users that can also edit images through Codex or Claude Code.

Pareto 26.9 routes requests across several frontier and open models and keeps the best answer, costing about one-third of Astra and finishing faster.

A user says Opus 5.5 shows better taste, intuition and vision than Astra, which needs heavy prompting and uses much more of the subscription allowance.

A blogger observes that Astra's token consumption spiked and then leveled off, while Opus 5.5 lowered spend on existing use cases, with overall ROI still rising.

A user asks OpenAI why Astra's safeguards prevent designing qPCR primers to detect pathogens like E. coli and Salmonella from public genomes.

A self-deprecating post jokes about setting Astra high goals and prompting persistently to churn out arXiv papers, plus buying OpenAI shares.

The author asked ChatGPT-6 Astra to create a single-file Three.js and WebGL Viking longship game with custom water shaders, PBR materials and oar splashes, at an estimated API cost of about $29.48.

The author says that after more than 400 hours on the same 3D scene, GPT-6 Astra keeps refining it instead of hitting a wall, which they cite as evidence it is close to AGI.

On the Artificial Analysis Index, Opus 5.5 leads with 58 points to Astra's 53, but it uses about 119k output tokens per task versus Astra's 27k and costs more per token.