GPT-6.1 Terra Makes OpenAI’s Midrange a Value Champion
Trending Topics Jakob Steinschaden ● Covered by 2 sources
OpenAI’s GPT-6.1 Terra is out, and it’s a lot cheaper than the last one. It also scores better and hallucinates far less, which is awkward for the pricier tiers.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has shipped GPT-6.1 Terra, its first Terra-class model since GPT-5.6. On paper, this is the middle rung in the company’s lineup: not the biggest model, not the cheapest one, but the place where price and performance are supposed to meet in the middle. This release is trying to make that middle look very attractive.
The pricing change is blunt. Input tokens are now $1 per million and output tokens $5 per million, down from $2 and $12 for GPT-5.6 Terra. Cached input falls to $0.05 from $0.20, while cache writes land at $1.25. Artificial Analysis says the model now delivers more intelligence for less than a third of the cost per task, with task cost dropping from $1.40 to $0.40 at maximum reasoning effort.
The gains are not just about price. In the Artificial Analysis Intelligence Index, GPT-6.1 Terra is up 6 points over GPT-5.6 Terra at maximum effort. At lower effort levels, the advantage grows to between 7 and 13 points. Against OpenAI’s own family, Terra’s highest setting sits 4 points behind GPT-6.1 Sol and 10 points ahead of GPT-6 Luna.
Reliability is where the model looks most improved. In AA-Omniscience, a private test for knowledge and hallucinations, Terra guesses when it does not know the answer in 38 percent of cases, down from 88 percent in the earlier version. Accuracy also rises by 2 points, so the model is both less prone to making things up and slightly better at getting answers right. That is a useful combination for anything that depends on trust, not just raw score chasing.
The coding numbers are solid, if not spectacular. On the new Artificial Analysis Coding Agent Index v2.0, Terra scores 48, behind Sol at 54 and ahead of Luna at 31. On Terminal-Bench 4.0, it gains about 20 points and now rivals GPT-6.1 Sol at maximum effort. Artificial Analysis also says low and medium effort settings are Pareto-optimal for cost efficiency, which is the bit developers and companies will actually care about when the bills arrive.
My take — AI-written commentary, not fact-checked reporting
OpenAI’s midrange finally looks like a product strategy instead of a placeholder. Cheaper, better, and less likely to hallucinate is the kind of upgrade that makes premium tiers sweat. The bigger story is that the market is forcing frontier labs to prove value, not just size, and that’s healthy for everyone except the marketing deck.
Read more about this at: Trending Topics
