Editor’s note (Oct 29, 2026): This article has been corrected. The original version ended with a research-process note stating that dedicated TechCrunch and The Verge coverage of Terra “was not located” – a statement about this desk’s retrieval, not about the world. That note has been removed, along with claims about missing specification details that were likewise descriptions of what our fetch returned rather than facts about the announcement. The article’s reporting, pricing, and availability facts are unchanged.
OpenAI announced GPT-6.1 Terra on October 26, 2026, introducing a new mid-sized model priced at $1 per million input tokens and $5 per million output tokens. This release arrives 27 days after the launch of GPT-6.1 Sol. For developers and operators, the model is positioned by OpenAI as a fast, efficient tool designed to improve visual reasoning and computer use capabilities while providing more natural writing performance. Before the announcement, TestingCatalog and kie.ai described Terra as an unannounced or leaked model surfaced in Microsoft evaluation configurations with unconfirmed pricing, a status now superseded by the official announcement.
Model positioning and availability
The official announcement frames GPT-6.1 Terra as a significant step up from the GPT-5.6 Terra model. OpenAI reports that the new model offers 50% lower standard API prices compared to its predecessor. Beyond the base rates, the model supports cached input at a price of $0.05 per million tokens, a feature intended to optimize costs for recurring data processing tasks. Availability is broad, covering Plus, Pro, Enterprise, and Edu users within ChatGPT Work and Codex. Developers can access the model via the OpenAI API using the identifier gpt-6.1-terra. According to ChatGPT release notes, the rollout also includes Business plans, though the model does not currently support UAE inference residency.
The evolving price ladder
At $1 per million input tokens and $5 per million output tokens, Terra is priced at exactly half the cost of the GPT-6.1 Sol model. The current price points for these models are: GPT-6 Luna at $0.10 per million input and $0.50 per million output, GPT-6.1 Terra at $1 per million input and $5 per million output, and GPT-6.1 Sol at $2 per million input and $10 per million output. This structure provides a granular selection of performance-to-cost ratios for different agentic workloads.
Cadence and family dynamics
The release of Terra marks the third distinct name to appear in the GPT-6.1 family within a four-week window, following the introduction of Sol and the subsequent press-reported cancellation of the Astra variant. The status of GPT-6.1 Astra as a cancelled project is based on secondhand, press-reported information, including Forkast’s own DevDay coverage, and does not appear on any OpenAI primary surface. The release notes indicate that some Plus and Pro users may see GPT-6.1 Terra at Medium instead of GPT-6.1 Sol at Light in default Power presets in ChatGPT Work. This shift in default routing demonstrates how quickly the company is reconfiguring its production environment to prioritize newer model deployments.
Vendor-reported performance metrics
OpenAI has provided several vendor-reported benchmarks to illustrate the capabilities of the new model. On the Agents’ Last Exam, the company claims that Terra substantially outperforms GPT-5.6 Terra in the context of software usage at work. Furthermore, on internal error-inducing prompts, the vendor-reported error rate for Terra is 16.7% at low reasoning, compared to 53.6% for GPT-5.6 Terra. At maximum effort, the error rate drops to 4.3%, which the company notes is an 18 percentage-point drop that brings performance nearly in line with GPT-6 Astra and GPT-6.1 Sol. These figures are OpenAI’s own measurements, reported here as such.
What remains unverified
The benchmark figures above have not been independently reproduced. Everything else reported here – price, cached-input rate, availability, the price ladder, and the release cadence – comes from OpenAI’s announcement and its published release notes.
