Terra handles visual reasoning, computer use, and writing, and it ships through the API as gpt-6.1-terra, in ChatGPT Work, and in Codex, though not in the standard chat interface. Cached input costs $0.05 per million tokens, a twentieth of the fresh input price, which rewards applications that resend the same instructions and tool definitions on every request. OpenAI says Terra substantially outperforms GPT-5.6 Terra on Agents' Last Exam, a benchmark for how well agents operate software at work, and completes tasks more than three times faster than both Astra and GPT-6.1 Sol.
Price Positioning
The GPT-5.6 family shipped in July split into three tiers, with Sol for frontier capability, Terra for balance, and Luna for high-volume work, and that structure already signalled that OpenAI intended to compete on cost per completed task rather than on benchmark scores. GPT-6.1 Terra collapses the middle tier into the frontier, because matching Astra at a tenth of the token price removes most of the reason to buy the expensive model for office work. A company that halves the price of the model directly beneath its flagship is not rationing capacity, it is buying volume, and it is telling buyers that the marginal cost of an agent doing white-collar work is now a fraction of the salary it replaces.
The Hit on Capex
Inference prices are now falling into a financing structure that assumes volume, and that structure is large. AI-linked debt issuance reached $466 billion this year, JPMorgan projects total AI debt above $4 trillion by 2030, and Nvidia arranged more than $500 billion of outside capital through Apollo, BlackRock, and Goldman Sachs against residual-value guarantees of up to 25%, including $105 billion for an OpenAI-linked project in Ohio. Pension funds and insurers bought that paper because the cash flows looked contracted and the compute looked scarce, and AI-related issuers now account for 49% of investment-grade corporate bonds. A 50% cut in price weakens the first assumption while doing nothing for the second, and the repricing channel runs through retirement accounts.
Breaking Value Chain
When prices get too cheap, it could collapse the whole industry revenue structure: insurers processing claims, banks reconciling breaks, logistics firms quoting freight, and back offices that currently pay per seat. At 10% of the token price and 3x the speed, work that failed the payback test a year ago clears it now, and the first budgets to move are the ones already spent on licensed software and outsourced staff. The losers are seat-based software vendors and business-process outsourcers whose pricing rested on per-human economics, while the compute owners sit in between, exposed to any fall in token volume that outruns the price decline.
Market Impact
The gap between token prices and token volumes would present itself as the most significant indicator rather than the benchmark scores. A model that is half the price and three times faster needs volume to grow faster than price falls just to hold revenue flat, and the buildout is financed on revenue rather than on capability. Two quarters of inference revenue growing through this cut would validate the debt, while volumes rising against flat revenue would turn the $4 trillion projection into a refinancing problem, and the price cut that made agents affordable would be the moment the arithmetic stopped working.
This marketing material is provided for informational purposes only and does not constitute investment advice, a recommendation, or an offer or solicitation to buy or sell any financial instruments.
Trading in securities involves significant risk and may not be suitable for all investors. Prices of securities may fluctuate significantly and may result in a total loss of your investment. Investors should be aware that losses may exceed potential profits when buying and selling securities. In certain market conditions, you may sustain losses that exceed your initial investment. Securities and contracts for differences are complex financial instruments that require a high level of knowledge and understanding. You should carefully consider whether you understand how these instruments work and whether you can afford to take the high risk of losing your money.
Source: https://developers.openai.com/api/docs/models/gpt-6.1-terra
