OpenAI’s GPT-6.1 Terra cuts API prices while improving benchmarks

OpenAI has released GPT-6.1 Terra, a mid-range model that The Decoder reports costs half as much through the API as GPT-5.6 Terra. The launch matters for developers and small teams because OpenAI is positioning Terra for everyday office work, automated software control and coding tasks where cost per run can decide whether an AI workflow is practical.

What OpenAI changed with Terra

According to OpenAI, GPT-6.1 Terra is designed to combine fast visual reasoning, more precise computer control and a more natural writing style. The company points to tasks such as drafting proposals, interpreting technical drawings and completing forms as intended use cases.

Terra is available now for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, as well as through the API.

The price shift is the headline for API users. The Decoder reports that GPT-6.1 Terra costs $1 per million input tokens and $5 per million output tokens. That compares with $2 and $12 for GPT-5.6 Terra. Cached input tokens fall to $0.05 per million, which The Decoder says is one quarter of the previous rate.

The same report frames Terra as OpenAI’s closest answer to Anthropic’s mid-tier Sonnet models. It lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens, while Terra’s listed prices are lower.

Why the price-performance claims matter

Artificial Analysis, cited by The Decoder, calls GPT-6.1 Terra a new high point for intelligence per dollar. At maximum reasoning effort, Terra improves by 6 points on the firm’s Intelligence Index compared with GPT-5.6 Terra, while the reported cost per task drops from $1.40 to $0.40.

At lower reasoning effort levels, Artificial Analysis reports larger gains of 7 to 13 points. It places Terra on the Pareto frontier, meaning the model sits at a point where, in that analysis, cost and performance cannot both be improved at the same time.

For readers building AI automations, that is the practical issue. Lower per-task cost can make it easier to run model calls in bulk, retry failed steps, process longer documents or add AI checks to routine workflows. If you are comparing automation stacks, our n8n vs Make vs Zapier guide is a useful companion to the model-pricing question, because orchestration costs and model costs both shape the final bill.

Coding agents and computer control are the focus

OpenAI is also presenting Terra as strong at automated software control. On the Agents’ Last Exam benchmark, The Decoder reports that Terra matches GPT-6 Astra’s performance at one-tenth the token price. Compared with Claude Opus 5.5 at maximum effort, Terra is reported to be more than three times as token-efficient at one quarter of the token price, which The Decoder says works out to about nine times lower cost per task.

OpenAI also says Terra completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. That speed claim is important for agentic workflows, where a model may need to inspect screens, decide next steps and interact with software repeatedly.

Coding results show a large improvement over the previous Terra model. On Terminal-Bench 4.0, The Decoder reports that GPT-6.1 Terra solves 50.3 percent of tasks at maximum reasoning effort, up from 21.5 percent for GPT-5.6 Terra. Artificial Analysis says this puts Terra in the same tier as GPT-6.1 Sol on that benchmark.

On Terminal-Bench-Science 0.1, Terra reportedly beats GPT-6.1 Sol at lower cost and comes within five points of Claude Opus 5.5 at less than one fifth of the estimated cost. OpenAI still recommends GPT-6.1 Sol and GPT-6 Astra for the most demanding coding work, so Terra is best read as a cost-efficient option rather than the top model for every programming job. For broader context on cheaper coding-agent competition, see our coverage of Deepseek V4.1 Flash 1017.

Fewer guesses, but safety remains part of the rollout

The Decoder reports a major reduction in hallucination behavior on Artificial Analysis’ private AA-Omniscience benchmark. When Terra lacks an answer, it guesses 38 percent of the time, down from 88 percent for GPT-5.6 Terra. Accuracy also increased by two points, according to the report.

For businesses, the guessing reduction may matter as much as raw benchmark gains. A cheaper model is only useful in production if it is less likely to fabricate when it should refuse, ask for clarification or disclose limits.

OpenAI says GPT-6.1 Terra meets the company’s “Critical” cybersecurity threshold under its Preparedness Framework and inherits GPT-6 Astra’s safeguards. The system card says Terra is more likely than its predecessor to stay inside assigned boundaries and disclose restrictions to users.

The system card also reports stronger defenses against indirect prompt injection, where malicious instructions are hidden inside third-party content. Terra’s defense rate is listed at 98.8 percent, up from 96.2 percent for GPT-5.6 Terra. That will not remove the need for review, permissions and monitoring, but it is a relevant metric for anyone considering Terra inside document, browser or desktop-control agents.