GPT-6.1 Terra halves OpenAI's API prices and still gains ground on its larger models

OpenAI has released GPT-6.1 Terra, a mid-range model the company describes as pushing the cost-to-performance boundary further than earlier versions. API pricing has been cut in half relative to its predecessor, GPT-5.6 Terra. The model is available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and through the API.
The price list is the headline. Input tokens cost $1 per million and output tokens $5 per million, against $2 and $10 for Anthropic's Claude Sonnet 5.5 and $2 and $12 for GPT-5.6 Terra. Cached input tokens fall to $0.05 per million, a quarter of the previous rate. Caching matters for applications where the same long instructions, documents or system prompts get reused across many calls, since that portion of the bill largely disappears. In practice, the gap between Terra and the Sonnet line is widest on generated text, which is often where chatty agent workloads spend most of their budget.
OpenAI frames the model around everyday work: drafting proposals, interpreting technical drawings, filling out forms, alongside precise computer control and what it calls a more natural writing style. That positioning suggests a model tuned for office and desktop automation rather than frontier reasoning.
Artificial Analysis, which tracks model prices and benchmark scores, calls GPT-6.1 Terra a new benchmark for intelligence per dollar. At maximum reasoning effort it gains six points on the group's Intelligence Index over GPT-5.6 Terra while costing less than a third per task, $0.40 versus $1.40. At lower effort levels the improvement widens to seven to 13 points, which Artificial Analysis places on the Pareto frontier, the point where cost and performance can no longer both be improved at once. On the Intelligence Index itself, Terra scores 48, sitting between GPT-6.1 Sol at 52 and GPT-5.6 Terra at 42.
Coding shows the largest single jump. On Terminal-Bench 4.0, Terra solves 50.3 percent of tasks at maximum reasoning effort, up from 21.5 percent for GPT-5.6 Terra. Artificial Analysis reads that roughly 20-point gain as putting the model in the same tier as GPT-6.1 Sol. On the newer Coding Agent Index v2.0, Terra scores 48, six points below GPT-6.1 Sol at 54 and 17 points above GPT-6 Luna at 31. On Terminal-Bench-Science 0.1 it beats GPT-6.1 Sol at lower cost and comes within five points of Claude Opus 5.5 while costing less than a fifth of the estimated price per task. OpenAI still points users to GPT-6.1 Sol and GPT-6 Astra for the most demanding coding work, so the cheaper model is not positioned as a full replacement at the frontier.
Computer control is the other area OpenAI emphasizes. On the Agents' Last Exam benchmark the model matches GPT-6 Astra at one-tenth the token price, and against Claude Opus 5.5 at maximum effort it is described as more than three times more token-efficient at a quarter of the token price, which works out to roughly nine times lower cost per task. OpenAI says it also finishes more than three times faster than both GPT-6 Astra and GPT-6.1 Sol. On OSWorld 2.1, a desktop agent benchmark, Terra beats GPT-5.6 Terra's best result at medium effort for a fraction of the cost. For visual reasoning, on MMMU-Pro it outperforms its predecessor at every reasoning level and nearly matches the larger GPT-6 Astra and GPT-6.1 Sol.
Hallucination behavior is a quieter but consequential change. On Artificial Analysis' private AA-Omniscience benchmark, the model guesses only 38 percent of the time when it lacks the answer, down from 88 percent for GPT-5.6 Terra, with accuracy up two points. A model that declines more often is easier to build reliable pipelines around, since wrong-but-confident answers are harder to catch downstream than refusals.
On safety, OpenAI classifies GPT-6.1 Terra as meeting the Critical cybersecurity threshold under its Preparedness Framework, and the model inherits GPT-6 Astra's safeguards. The system card reports it holds up about as well as GPT-6 Astra and GPT-6.1 Sol against direct attacks on the instruction hierarchy, with those models reaching 99.99 percent on such tests. Against indirect prompt injection, where malicious instructions are hidden in third-party content, Terra's defense rate is 98.8 percent, up from 96.2 percent for GPT-5.6 Terra. That category deserves attention from anyone shipping agents that read web pages, email or uploaded files, though a higher defense rate is not immunity.
Why it matters: Teams paying for mid-tier inference now face a substantial price cut at the same tier, and cached input at $0.05 per million makes repeat-prompt workloads like document processing and desktop agents radically cheaper to run. The benchmark numbers suggest a capable default for day-to-day tasks, with larger models reserved for frontier coding and reasoning. My inference is that the biggest near-term effect will be on routing decisions: products that previously escalated to a flagship model for computer control or moderate coding may now be able to keep more traffic on the cheaper tier, while prompts that need the strongest reasoning still go elsewhere.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
OpenAI has released GPT-6.1 Terra, a mid-range model that cuts API prices in half compared to its predecessor. Analysts say it sets a new standard for intelligence per dollar, especially for everyday computing tasks and AI-driven computer control. The article GPT-6.1 Terra halves OpenAI's API prices and still gains ground on its larger models appeared first on The Decoder .