GPT-6.1 Terra Delivers Double the Terminal-Bench Performance at Half the Price
GPT-6.1 Terra has been officially integrated into GitHub Copilot. OpenAI's new mid-tier model cuts API pricing to half that of its predecessor while more than doubling its Terminal-Bench score (a test measuring how well a model handles tasks in a command-line environment).
According to GitHub, GPT-6.1 Terra is available in the model picker for Copilot Pro, Pro+, Max, Business, and Enterprise users. Its supported environments are broad, spanning Visual Studio Code, Copilot CLI, Copilot cloud agents, the Copilot app, github.com, GitHub Mobile for iOS and Android, JetBrains IDEs, Xcode, and Eclipse.
GPT-6.1 Terra is priced at $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million tokens. That is a quarter of the previous rate, and lower than Claude Sonnet 5.5 ($2 input / $10 output) and GPT-5.6 Terra ($2 input / $12 output). Within OpenAI's lineup, it costs half as much as GPT-6.1 Sol and GPT-5.6 Terra, and roughly 10 times as much as GPT-6 Luna.
The enterprise rollout process is also set. Admins on Enterprise and Business plans control whether GPT-6.1 Terra is available through model policies, and it is enabled by default unless explicitly turned off. Usage is billed at the provider's official list price based on consumption. OpenAI recommends using GPT-6.1 Sol or GPT-6 Astra for the most demanding coding tasks.
The performance gains are backed by the numbers. On Terminal-Bench 4.0 at maximum reasoning effort, the model scored 50.3%, surpassing GPT-5.6 Terra's 21.5%. OpenAI said that on its internal benchmark, the error rate at low reasoning effort dropped from 53.6% to 16.7%, and fell to 4.3% at maximum reasoning effort. On Artificial Analysis's Coding Agent Index v2.0, it scored 48 — six points below GPT-6.1 Sol (54) and 17 points above GPT-6 Luna (31). OpenAI also said it completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.
Hallucination improvements were also confirmed. On the AA-Omniscience benchmark, the share of responses where the model guessed instead of admitting it didn't know dropped from GPT-5.6 Terra's 88% to 38%. GPT-6.1 Terra is positioned as the mid-tier model in OpenAI's lineup, ordered by price and performance: Luna, Terra, Sol, and Astra.
