OpenAI releases GPT-6.1 Terra mid-range model at $1/$5 per million tokens, half the price of GPT-5.6 Terra

OpenAI released GPT-6.1 Terra, a mid-tier model, to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and through the API. API pricing is $1 per million input tokens and $5 per million output tokens, against $2 and $12 for GPT-5.6 Terra according to the article. Cached input is $0.05 per million. Artificial Analysis reports an Intelligence Index of 48 (up from 42), roughly $0.40 per task at max effort (down from $1.40), and Terminal-Bench 4.0 of 50.3% (up from 21.5%). OpenAI still recommends GPT-6.1 Sol and GPT-6 Astra for the hardest coding work.

Developers running high-volume workloads on the previous Terra model have a clear reason to test a swap, since both the list price and the reported cost per task fall. Note that the output price drop is about 58% and the input drop is 50%, so "half the price" is roughly right but not exact. Rivals, Anthropic's Sonnet tier in particular, face pressure to respond on price. The effect is limited to mid-range use cases, because OpenAI itself points the most demanding coding tasks to its larger models.

"Pushes the cost-to-performance frontier further than any predecessor" and "a new benchmark for intelligence per dollar" are framing from OpenAI and a single benchmark firm, not settled fact. The computer-control and fewer-hallucination claims rest mainly on OpenAI's own figures, and the extraction found no primary links to check them. The jump from 21.5% to 50.3% on Terminal-Bench 4.0 and the fall in guessing rate from 88% to 38% are large and deserve independent replication before anyone treats them as typical.

1. 01Independent replication of the Terminal-Bench 4.0 and AA-Omniscience results by other evaluators, and whether they match Artificial Analysis within a few points.
2. 02Anthropic's response in the next few weeks: any price change or new release in the Claude Sonnet tier relative to the $1/$5 price point.
3. 03Real-world cost per task from developers on production workloads, including whether the reported $0.40 per task holds outside max-effort benchmark settings and whether rate limits or availability issues appear.