AI Skill of the Day: Race two models on the same boring task
Nerd's Chalk ● Covered by 8 sources
OpenAI rolled out GPT-6.1 Terra to ChatGPT Work, Codex and the API. It’s built for visual work and cheaper than Sol, so the boring tasks just got faster and less pricey.
Based on reporting by Nerd's Chalk — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI started rolling out GPT-6.1 Terra on October 26, 2026, and the pitch is simple: faster work on images, interfaces and writing without paying top-tier prices. Eligible ChatGPT Work and Codex subscribers get it first, and developers can call it through the API. Free-tier access is not supported.
Terra is meant for the kind of stuff people actually do all day. OpenAI points to technical drawings, proposals, forms, video editing, product photos and document-based revisions. In its own guidance, the model sits in the everyday-work slot: not the most powerful model in the stack, but the one you reach for when you need something clear, useful and quick.
The price drop is the part that will make finance people blink. Standard API input is $1 per million tokens and output is $5 per million, which OpenAI says is half the rate of GPT-6.1 Sol. Compared with GPT-5.6 Terra, the newer model is cheaper on both input and output, and it gets a bigger Codex context window too: 500,000 tokens versus 272,000.
OpenAI’s own benchmarks back up the cost story. On Agents’ Last Exam, Terra scored 57.6% at medium effort for $0.29 per task, while GPT-5.6 Terra scored 42.6% for $0.84. It matched GPT-6 Astra’s medium-effort score at a much lower task cost, and at max effort it finished in 60.0% for $1.01 per task. OpenAI also says it takes less than a third of the time required by GPT-6 Astra and GPT-6.1 Sol in its measurements.
Nerdschalk’s Codex tests tell the same story in miniature. Using Codex CLI 0.159.0 on a Linux server, Terra got every one of nine runs correct across three tasks and posted the shortest median time on each one. It finished the three-task set in 27.6 seconds, compared with 49.7 seconds for Sol. That’s a small test, sure, but it’s the kind of boring, repetitive work where a model either saves time or becomes another tab nobody trusts.
My take — AI-written commentary, not fact-checked reporting
This is the AI market at its most honest: not grand claims, just cheaper tokens and fewer mistakes on dull jobs. OpenAI is clearly pushing a split between prestige models and workhorse models, which is sensible for once. The industry could use more of this and less mystical wallpaper about intelligence going brrr.
Read more about this at: Nerd's Chalk