Originally published on The Daily Flare.
OpenAI released GPT-6.1 Terra on Oct. 26, setting API prices at $1 per million input tokens and $5 per million output tokens, with access in ChatGPT Work and Codex.
The release upgrades OpenAI’s Terra model line from GPT-5.6 Terra. FeaturedDaily reported on Nov. 9 that the new model was also available through the API, while its rollout in ChatGPT was continuing for Plus, Pro, Enterprise and Edu subscribers.
Terra is the second of four levels in OpenAI’s model lineup, below GPT-6.1 Sol and GPT-6 Astra and above Luna, according to FeaturedDaily. The company positions it as a faster, lower-cost option for work involving visual reasoning and computer use, rather than its top choice for the most demanding coding jobs.
The API rates are half those of GPT-6.1 Sol, which costs $2 per million input tokens and $10 per million output tokens. GPT-6 Astra costs $10 and $50, respectively, making Terra one-tenth its price per token, according to figures reported by OKdiario. The previous GPT-5.6 Terra was priced at $2 per million input tokens and $12 per million output tokens.
OpenAI says the new model completes tasks more than three times as quickly as GPT-6 Astra and GPT-6.1 Sol. The company also says Terra can perform on par with Astra on some tasks at one-tenth the price per token. Those are OpenAI’s comparisons, not results from an independent evaluation.
Benchmark gains and limits
On Agents’ Last Exam, Terra scored 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort, at about one-third of the cost, FeaturedDaily reported. The test includes professional tasks such as turning architectural plans into a 3D model and reviewing animation footage to create a time-coded record.
On Terminal-Bench 4.0, Terra scored 50.3% at maximum reasoning effort, compared with 21.5% for GPT-5.6 Terra. The benchmark result was cited in reporting by FeaturedDaily and OKdiario. The figures indicate improvement on that test, but do not establish how the model will perform across every coding or computer-use task.
OpenAI also reported a lower error rate on an internal benchmark of prompts designed to induce mistakes. At low reasoning effort, the rate fell from 53.6% for GPT-5.6 Terra to 16.7% for GPT-6.1 Terra. Because the benchmark is internal to OpenAI, the result has not been presented as an independent test. Artificial Analysis separately reported a substantial reduction in hallucinations on its own benchmark, without a specific figure in FeaturedDaily’s account.
OpenAI describes Terra as combining visual reasoning with computer control. OKdiario reported that the model can take text and images as input and produce text, but does not accept audio or video as direct inputs. The outlet also reported a context window of 1.05 million tokens, a maximum response length of 128,000 tokens and a knowledge cutoff of June 1, 2026.
Potential uses cited in OKdiario’s report include drafting proposals, interpreting technical plans, filling in forms and editing video. For businesses, the outlet described processing invoices and entering their details into accounting software as one example. The model can also match product photographs with catalog entries and interact with application screens.
OpenAI continues to recommend GPT-6.1 Sol or GPT-6 Astra for the most complex coding tasks, according to FeaturedDaily. That distinction leaves Terra aimed at workloads where speed and lower token costs matter, while reserving the higher-tier models for harder work.
TechTarget Japan’s Nov. 9 analysis noted that falling model prices can make it harder for companies to decide which system suits each workload. Futurum Group analyst David Nicholson said businesses still need to match tasks to models to find cost-effective choices. Gartner analyst Arun Chandrasekaran said faster agent-driven workflows could benefit from Terra, while stressing that scaling autonomous systems requires controls such as task-level return-on-investment measurement, access restrictions and human oversight.
Top comments (0)