OpenAI's latest advancements in its GPT series have brought significant improvements to the multimodal large language model, GPT-5. The company has released several updates and evaluations of the model, showcasing its enhanced capabilities in various tasks, including legal reasoning, organizational structure, and audience calibration.
What Happened
In a recent series of releases, OpenAI has shared updates on its GPT-5 model, which was launched in August 2025. The company has provided evaluations of the model's performance in various tasks, including legal reasoning, organizational structure, and audience calibration. One of the key updates is the release of GPT-5.6, a preview system card that assesses the model's monitorability using chain-of-thought (CoT) measures.
The CoT monitorability evaluates whether a monitor can infer particular properties of a model's behavior from its reasoning trace. This measure has been shown to provide a substantially richer monitoring signal than actions and final outputs alone, according to OpenAI. The company has also introduced three new monitorability environments based on other system card safety evaluations: destructive actions, confirmation consent, and background work.
These evaluations demonstrate the model's ability to handle complex tasks, such as risk assessment, deal management, and analysis of litigation filings. GPT-5.6 has shown significant improvements in these areas, with a score of 91.7% on the BigLaw Bench evaluation suite, up from GPT-5.4's 91.0%. The model also achieved 43% perfect scores and 87% of tasks scored above 0.80.
Background and Context
GPT-5 is a multimodal large language model developed by OpenAI, the fifth in its series of generative pre-trained transformer (GPT) foundation models. The model was launched on August 7, 2025, and is publicly accessible to users of the chatbot products ChatGPT and Microsoft Copilot as well as to developers through the OpenAI API.
The GPT series has been designed to improve upon previous models in various areas, including legal reasoning, organizational structure, and audience calibration. The latest updates demonstrate the company's continued efforts to enhance the model's capabilities and address challenges such as verbosity, which remains an occasional challenge for the model.
Why it Matters to the Industry
The advancements in GPT-5 have significant implications for industries that rely on language processing, including the adult industry. The model's improved performance in tasks such as risk assessment and deal management can be applied to various areas, such as age verification and payment processing.
Moreover, the model's enhanced organizational structure and audience calibration capabilities can improve content moderation and filtering systems, reducing the burden on human moderators and improving the overall user experience. The increased accuracy and precision of the model also reduce the risk of false positives or negatives in age-gating and content classification tasks.
What Comes Next
OpenAI has announced that GPT-5 is currently in research preview, with plans to roll out the model to eligible clients once it becomes generally available. The company will continue to evaluate and refine the model's performance in various tasks, addressing challenges such as verbosity and improving its overall capabilities.
Key Facts
- GPT-5 is a multimodal large language model developed by OpenAI, launched on August 7, 2025.
- The model has shown significant improvements in tasks such as legal reasoning, organizational structure, and audience calibration.
- OpenAI has introduced three new monitorability environments based on other system card safety evaluations: destructive actions, confirmation consent, and background work.
- GPT-5.6 has scored 91.7% on the BigLaw Bench evaluation suite, up from GPT-5.4's 91.0%.
- The model achieved 43% perfect scores and 87% of tasks scored above 0.80.
- GPT-5 is currently in research preview, with plans to roll out the model to eligible clients once it becomes generally available.
