OpenAI's GPT-6.1 Terra: half the price, now in Codex and ChatGPT
The mid-tier model left out of September's GPT-6 launch lands with 1.1M context, big coding and computer-use gains, and near-Astra quality at a tenth the cost.
Copy markdown
Half the price of the model it replaces
GPT-6.1 Terra ships at $1 / $5 per million input/output tokens (cache reads just $0.05/M) — 50% below GPT-5.6 Terra and GPT-6.1 Sol — with a 1.1M-token context. It's live now via the API (gpt-6.1-terra), ChatGPT Work, and Codex. On OpenRouter, a Flex tier runs cheaper still at $0.50 / $2.50.
Near-frontier quality at a tenth the cost
On Agents' Last Exam, Terra matches GPT-6 Astra at roughly one-tenth of Astra's per-token price, and OpenAI pegs its task cost at about 9x lower than Claude Opus 5.5 at max reasoning effort. If you've been paying flagship rates for mid-tier jobs, the math just flipped.
A real coding jump — but not for the hard stuff
Terminal-Bench 4.0 climbs to 50.3% at max effort, up from 21.5% for GPT-5.6 Terra, and OpenAI says Terra finishes tasks 3x faster than Astra and Sol. The caveat: OpenAI still points you to Sol and Astra for the most complex coding, so treat Terra as the fast default, not the heavy lifter.
Fewer hallucinations, better computer use
On OpenAI's internal factuality test, error rate drops from 53.6% to 16.7% at low effort and 4.3% at max. Terra also gains on OSWorld 2.1 (desktop automation) and MMMU-Pro (visual reasoning) — useful if your agents click through real UIs. One flag: prep evals rate it Critical for cybersecurity.
Also today: Devin Security 2.0 ships fixes as PRs
Cognition launched an agentic security workflow that scans repos with Claude Mythos and other cyber models, triages findings against your threat model, then spins up a VM per issue to reproduce, patch, test, and open a pull request. It's enterprise-gated, but Cognition is offering free scans and remediation credits to new customers.