OpenAI shipped GPT-6.1 Terra on Monday, a mid-tier refresh aimed squarely at the high-volume computer work that fills an office day — reading invoices, filling forms, drafting proposals, cutdown video edits. The headline number is price: standard API pricing of US$1 per million input tokens and US$5 per million output tokens, half of what GPT-5.6 Terra charged, with cached input dropping to a quarter of the old rate. As Quartz reports, the model is available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and via the API as gpt-6.1-terra.
The benchmarks OpenAI published put Terra in unusual territory: near-frontier results at mid-tier prices. On Agents’ Last Exam, a benchmark of professional computer tasks, Terra scores 15 percentage points above GPT-5.6 Terra at medium reasoning effort while costing about a third as much, and matches GPT-6 Astra — OpenAI’s most capable model — at one-tenth of the per-token cost, all while running more than three times faster than Astra or GPT-6.1 Sol. At maximum effort, OpenAI claims Terra is over three times as token-efficient as Anthropic’s Claude Opus 5.5 at a quarter of the price per token, working out to roughly a ninth of the cost per task — figures OpenAI acknowledges rest on its own internal evaluations, with no outside party having confirmed them.
What independent testing says — and what it does not
Independent lab Artificial Analysis corroborates the broad shape: their Intelligence Index scores Terra at maximum effort 6 points above GPT-5.6 Terra at less than a third of the cost per task (US$0.40 versus US$1.40), though they find only the low- and medium-effort settings sit on the cost-efficiency frontier — max effort falls behind it, and the model lands 4 points below GPT-6.1 Sol. Their own hallucination benchmark shows a dramatic drop: Terra guesses 38% of the time it does not know an answer, down from 88% for GPT-5.6 Terra, while accuracy slightly improved. On Terminal-Bench 4.0, the command-line coding test, Terra solves 50.3% of tasks at maximum effort — more than double its predecessor’s 21.5%. For the hardest coding, OpenAI still points users to Sol or Astra.
The Register’s verdict was measured — “not a bombshell, but it rivals pricier models on some tasks at a tenth of Astra’s API price,” as their launch coverage put it. That is a fair summary of OpenAI’s positioning: this is a price story more than a capability story, the continuation of a race we have tracked since OpenAI slashed GPT-5.6 prices by up to 80% in July and since GPT-6 Astra’s launch brought limited, tiered access to the frontier model. The four-tier ladder — Luna, Terra, Sol, Astra — now has a refreshed middle rung.
The Critical cyber rating nobody is buying
One detail in the launch will outlast the price cuts: GPT-6.1 Terra meets the Critical threshold in cybersecurity under OpenAI’s Preparedness Framework, inheriting Astra’s protections. That is the same threshold OpenAI publicly overrode for GPT-6 Astra’s release in September — the first time the company shipped a model past its own declared red line. Now the Critical rating has trickled down to a US$1-per-token mid-tier model, meaning high-effort agentic computer-work automation at commodity prices includes frontier-class cyber capability by default.
It cuts both ways in the efficiency-vs-risk trade OpenAI is making. On the risk side, the same cost collapse makes broad agentic deployment affordable for organisations with no security team — a point made sharper by the CPU-cache covert-channel attack demonstrated against AI agents nine days ago, which showed the agent layer itself still has unpatched attack surfaces. On the commercial side, the collapse of per-task cost to cents is what finally makes “AI does the paperwork” stack up for organisations that never had a margin for it.
The NZ angle: cents per task changes the maths for small operators
New Zealand buyers feel API prices in full USD plus exchange rates, which is precisely why the mid-tier matters more here than in California. Cost per task at Terra’s rates lands at a few cents for routine document work — low enough that a five-person Otago accounting practice or a Tauranga conveyancing shop can automate invoice-processing without a business case, the way they once automated spreadsheets. The constraint is no longer API spend; it is trust and verification — knowing what the agent did, and being able to check. OpenAI’s price cut moves the frontier debate in the direction of who can audit a $0.04 mistake, because the $9-per-task era at least priced in human review.
