OpenAI Unveils GPT-6.1 Terra with 50% Price Cut and Enhanced Performance
OpenAI releases GPT-6.1 Terra, halving API pricing while delivering strong benchmark results across coding, computer control, and visual reasoning tasks.
OpenAI has officially launched GPT-6.1 Terra, a new mid-tier model designed to deliver a dramatic improvement in cost efficiency while rivaling significantly larger frontier models. Positioned directly against Anthropic's Claude Sonnet series, Terra cuts API pricing in half compared to its predecessor, GPT-5.6 Terra. The model costs $1 per million input tokens and $5 per million output tokens, alongside a reduced rate of $0.05 per million cached input tokens. By comparison, Claude Sonnet 5.5 stands at $2 for inputs and $10 for outputs, while GPT-5.6 Terra charged $2 and $12 respectively. Terra is available across ChatGPT Work, Codex, and OpenAI's API for Plus, Pro, Enterprise, and Edu tier subscribers.
Performance gains and cost efficiency
According to evaluations by Artificial Analysis, GPT-6.1 Terra establishes a new standard for intelligence per dollar, placing the model on the Pareto frontier where performance and cost gains converge. On the Intelligence Index, Terra achieves a score of 48, placing it between GPT-5.6 Terra's score of 42 and GPT-6.1 Sol's score of 52. At maximum reasoning effort, Terra gains 6 points over GPT-5.6 Terra while reducing per-task costs from $1.40 down to $0.40. At lower reasoning effort levels, performance gains reach between 7 and 13 points.
Beyond general reasoning, Terra demonstrates marked reductions in hallucinations. On the private AA-Omniscience benchmark, the model's reliance on guessing when lacking an answer dropped to 38 percent, down from 88 percent in GPT-5.6 Terra, alongside a two-point improvement in overall accuracy. On the Coding Agent Index v2.0, Terra scored 48 points, putting it 17 points ahead of GPT-6 Luna's score of 31 and just 6 points behind GPT-6.1 Sol's score of 54.
High-speed computer control and developer benchmarks
OpenAI engineered GPT-6.1 Terra with a strong emphasis on automated software manipulation, visual reasoning, and coding tasks. In computer control tests on the Agents' Last Exam benchmark, Terra matches the performance of GPT-6 Astra at one-tenth of the token price. When evaluated against Claude Opus 5.5 at maximum reasoning effort, Terra proves over three times as token-efficient at one-fourth the token cost, translating to roughly nine times lower cost per completed task. OpenAI also reports that Terra executes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol.
On technical benchmarks, Terra solves 50.3 percent of tasks on Terminal-Bench 4.0 at maximum reasoning effort—up from 21.5 percent with GPT-5.6 Terra—bringing it into the same operational tier as GPT-6.1 Sol. For scientific coding on Terminal-Bench-Science 0.1, Terra outperforms GPT-6.1 Sol at lower expense and comes within five points of Claude Opus 5.5 at less than a fifth of the estimated cost. In visual reasoning via MMMU-Pro, Terra matches the performance levels of GPT-6 Astra and GPT-6.1 Sol across reasoning stages, while also beating GPT-5.6 Terra on the OSWorld 2.1 desktop agent benchmark.
Enhanced safety standards
From a security perspective, GPT-6.1 Terra meets OpenAI's "Critical" cybersecurity threshold under its Preparedness Framework, inheriting safety measures established in GPT-6 Astra. The system card indicates that Terra achieves a 99.99 percent defense rate against direct attacks on instruction hierarchy, matching GPT-6 Astra and GPT-6.1 Sol. For indirect prompt injection attacks, where malicious commands are embedded in third-party content, Terra successfully defends against 98.8 percent of attempts, an improvement over GPT-5.6 Terra's 96.2 percent mark.
What it means for developers
The arrival of GPT-6.1 Terra provides software engineers with a viable high-performance model for routine agentic workflows, desktop automation, and code execution without incurring high operational costs. OpenAI still recommends flagship models such as GPT-6.1 Sol and GPT-6 Astra for extremely complex coding tasks, but Terra offers an attractive alternative for automated form filling, proposal drafting, and technical drawing analysis.
For engineering teams evaluating model options across providers, developers can try top AI models cheaply through one API at https://apixoai.online. With Terra's cached inputs dropping to $0.05 per million tokens, building high-frequency applications with persistent context becomes significantly more economical. The combination of faster processing speeds, lower hallucination rates, and strong security guardrails allows developers to deploy reliable autonomous agents at scale.
Source: GPT-6.1 Terra halves OpenAI's API prices and still gains ground on its larger models — The Decoder. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
