Technology News

GPT-6.1 Terra is available: OpenAI improves ChatGPT and Codex

GPT 61 Terra is available OpenAI improves ChatGPT and Codex.jpg

Open AI advertisement GPT-6.1 Terra, a new version of its intermediate AI model that combines visual reasoning, the use of applications and text generation at a lower cost. Available today on ChatGPT Work, Codex, and Developer API, this model promises advancements in programming and a 50% drop in token prices compared to its predecessor.

ChatGPT logo

GPT-6.1 Terra succeeds the GPT-5.6 Terra in the OpenAI model range. The company presents it as an AI model capable of managing common tasks in a professional environment: writing proposals, interpreting technical plans, filling out forms or even modifying videos. It combines image analysis and the use of computer interfaces with the aim of automating operations that until now required several human manipulations.

This ability to act in applications constitutes one of the main axes of the new version. For example, an agent can review invoices and then enter their information into accounting software. OpenAI also highlights complex tasks that are carried out in multiple steps, such as creating a 3D model from an architectural drawing.

Halving the price of GPT-6.1 Terra and fewer factual errors

OpenAI announces a 50% reduction in the standard API price compared to GPT-5.6 Terra, but also GPT-6.1 Sol. The new model costs $1 per million input tokens, $0.05 for cached input tokens, and $5 per million output tokens. This price should make it easier to use for large scale repetitive operations.

The company also claims to have improved reliability. In its internal benchmark dedicated to indications that can cause errors, the error rate ranges from 53.6% for GPT-5.6 Terra to 16.7% with GPT-6.1 Terra when reasoning effort is low. At the maximum level it reaches 4.3%, a drop of 18 points. These figures do not guarantee the absence of hallucinations in real situations, but they indicate advances in the tests used by OpenAI.

Programming is also progressing: at the maximum level of reasoning, GPT-6.1 Terra solves 50.3% of Terminal-Bench 4.0 benchmark tasks, compared to 21.5% for GPT-5.6 Terra according to OpenAI results. However, the company specifies that its most powerful models, GPT-6.1 Sol and GPT-6 Astra, are still preferable for more complex development tasks.

GPT-6.1 Terra is now available for ChatGPT Plus and Pro subscribers, as well as Enterprise and Edu users on ChatGPT Work and Codex. Developers can also integrate it into their applications via the API, with the identifier gpt-6.1-terra.

Shares:

Related Posts