Terra is available now to Plus, Pro, Enterprise, and Edu users inside ChatGPT Work and Codex. Developers can reach it through the API as gpt-6.1-terra.
Half the Price, Faster, and Far Fewer Errors
API pricing is 50 percent lower than for GPT-5.6 Terra and GPT-6.1 Sol. Tokens are the segments of text a model processes, and Terra is billed per million of them:
- Input tokens: $1
- Cached input tokens: $0.05
- Output tokens: $5
On Agents’ Last Exam, a benchmark for professional tasks, Terra scored 15 percentage points higher than GPT-5.6 Terra at one-third of the cost. OpenAI says it performs on par with GPT-6 Astra at one-tenth the price per token, and finishes tasks more than three times faster than both Astra and Sol.
Against rivals, the claims get sharper. At maximum effort, Terra is over three times more token efficient than Claude Opus 5.5 and costs one-fourth as much per token, which works out to a nine times lower cost per task. The model also held up well in prompt injection tests.
Reliability improved too. On OpenAI’s internal benchmark of error-inducing prompts, Terra’s error rate at low reasoning effort fell from 53.6 percent with GPT-5.6 Terra to 16.7 percent, or roughly one prompt in six. At maximum effort it dropped 18 percentage points to 4.3 percent.
Coding is where the discount shows. On Terminal-Bench 4.0, which evaluates coding agents, Terra solved 50.3 percent of tasks at maximum effort, up from 21.5 percent for GPT-5.6 Terra. OpenAI still points users to GPT-6.1 Sol and GPT-6 Astra for the hardest coding problems. For research tasks that involve code, though, Terra beats Sol at lower cost and lands within five points of Opus 5.5 at less than one-fifth of its estimated price.
What “Critical” Means Under OpenAI’s Framework
Under its April 2025 Preparedness Framework, OpenAI says Terra reaches the “Critical” cybersecurity level. In practice, that means a model that, given tools, can find and build exploits for previously unknown software flaws, known as zero-days, and can plan and carry out new attacks against hardened targets without human help. OpenAI’s own rules require safeguards during development for a model at this level.
Hashlytics Take
The interesting part is the pairing. A model OpenAI classifies as capable of autonomous exploit development is also the one priced for volume, at a fraction of Astra’s cost per token. Cheap tokens widen who can afford to run it, and the only gate between that capability and misuse is the safeguards OpenAI says it built. The security case for enterprises also leans heavily on OpenAI’s own evaluations, with prompt injection resistance reported in a single line rather than with numbers. IT teams weighing Terra for core operations should read the cost savings as the easy part of the pitch and ask to see the safeguards in detail before the rollout.
Follow Hashlytics on Bluesky, Facebook, LinkedIn , Telegram and X to Get Instant Updates
