OpenAI announces GPT-4.5, its largest and most knowledgeable model yet

Today, OpenAI announced its new non-reasoning model, GPT-4.5, which is the largest and most knowledgeable model to date. As the name suggests, GPT-4.5 is built upon GPT-4o by further scaling it during the pre-training process. OpenAI has confirmed that GPT-4.5 is not a frontier model, but it is their largest LLM, boasting more world knowledge, improved writing skills, and a refined personality compared to GPT-4o.

Benchmark data indicates that GPT-4.5 is a modest upgrade from GPT-4o. On the SWE-bench Verified benchmark, GPT-4.5 achieved 38%, an uplift of 2-7% over GPT-4o and 30% below OpenAI’s O3-based deep research. For comparison, Anthropic’s Claude 3.7 Sonnet achieved a state-of-the-art performance of 62.3% on SWE-bench Verified.

On the SimpleQA Accuracy benchmark, GPT-4.5 scored 62.5%, compared to 38.2% for GPT-4o. On the SimpleQA Hallucination Rate benchmark, GPT-4.5 scored the lowest among all of OpenAI's large language models.

Recently, OpenAI’s Preparedness team developed a new benchmark called SWE-Lancer, which evaluates LLM performance on real-world software engineering tasks, including feature development, design, bug fixes, and more. In this new benchmark, the GPT-4.5 model was able to solve 20% of IC SWE tasks and 44% of SWE Manager tasks, a slight improvement over OpenAI’s o1 model. You can read the new model’s full System Card here.

On the safety front, based on the results from the Preparedness evaluations, OpenAI’s Safety Advisory Group has classified this new GPT-4.5 model as medium risk overall. It also scored low for cybersecurity and model autonomy.

The new GPT-4.5 model research preview is now available for ChatGPT Pro users via the model picker on web, mobile, and desktop. It supports search, file and image uploads, and canvas features on ChatGPT. Multimodal features like Voice Mode, video, and screen sharing will be supported in the future. GPT-4.5 will be available next week to ChatGPT Plus and Teams users.

GPT-4.5 is also available for all paid-tier developers via the Chat Completions API, Assistants API, and Batch API. It supports key features such as function calling, Structured Outputs, Streaming, and system messages.