Summary
Nick Saraev puts OpenAI’s GPT 6.1 Terra head to head with Anthropic’s Claude Opus 5.5 on practical build tasks rather than standard benchmarks. Both models received identical prompts and tools, running at max effort in their own harnesses, Codex for Terra and Claude Code for Opus.
On the numbers, Terra averaged about 23 minutes 43 seconds per build versus 43 minutes 52 seconds for Opus 5.5, used roughly a third of the output tokens, and costs a small fraction of Opus, which the presenter estimates at under one-twelfth. The demos include a live storm-tracking globe, a stormy lighthouse simulation, a Minecraft-style builder, a racing game, and a luxury watch landing page.
The takeaway is that Terra performs impressively on 2D and simpler projects at a much lower price, while Opus 5.5 pulls clearly ahead on complex 3D work and high-fidelity design. The video also looks at Terra’s reasoning effort options, ranging from light to ultra, and notes that its speed and cost make it attractive for computer-use tasks.
It is a useful look at where a cheaper model is good enough and where a premium model still earns its price.
📺 Source: Nick Saraev · Published October 30, 2026
🏷️ Format: Comparison
