Summary
Nate Herk puts OpenAI’s GPT-6.2 Astra head-to-head with Anthropic’s Claude Fable 5.2 across 12 real-world use cases, including web design, slide decks, travel planning, 3D games, financial dashboards and spreadsheets, and editing Instagram and YouTube videos. Both models run at high effort in separate work trees with identical prompts, and each test reports dollar cost, input and output tokens, and runtime.
The first round is a blind website comparison for the fictional Perk Form brand. Fable ran for 56 minutes at roughly $175 in API-equivalent cost, while Astra finished in 12 minutes for about $2.50. Herk picks Astra’s cleaner site, and the scoreboard moves back and forth over later rounds, with Fable winning on tasks such as a virtual event experience and the Instagram reels sound design.
A key takeaway concerns pricing. Both models are listed at $8 per million input tokens and $40 per million output tokens, yet Astra and its Codex harness use far fewer tokens, which makes it much cheaper in practice. Viewers looking to decide which model suits a given kind of work will find concrete outputs, costs and verdicts for each category.
📺 Source: Nate Herk | AI Automation · Published November 18, 2026
🏷️ Format: Comparison







