OpenAI shared the benchmark scores of the o3 series of artificial intelligence (AI) models on Friday. As successor to the recently released o1 series of AI models, the company claimed that o3 significantly surpasses the capabilities of the older version. The series has two models, the o3 and o3-mini, and is focused on advanced reasoning-based tasks. Currently, the AI models are being made available for public safety testing, and the o3-mini model will be launched in the public domain early next year. Notably, OpenAI’s announcement came just days after Google released its Gemini 2.0 Flash Thinking Mode AI model.
Table of Contents
OpenAI Shares Benchmark Scores of the o3 Series AI Models
On day 12 of the AI firm’s 12-day shipping schedule, OpenAI announced the o3 series, the successor to the o1 AI model. During a live stream, CEO Sam Altman highlighted that the name o2 was dropped due to the UK telecommunications service provider Telefonica which uses the same brand name.
OpenAI highlighted that despite the announcement, the models will currently not be available in the public domain. The company has started providing both o3 and o3-mini in early access to selected external researchers for public safety testing. Those interested in participating in the process can apply here. The application process ends on January 10.
The company highlighted that both the o3 series AI models will be more powerful than their predecessors, and will be able to perform more complex tasks in coding, mathematics, and natural language processing. However, its full range of capabilities was not revealed.
Additionally, OpenAI shared some benchmark evaluations of the o3 AI models during internal testing. The company claimed that o3 scored 71.7 percent in the SWE-bench benchmark and 96.7 percent in the AIME 2024 benchmark, surpassing o1. However, a full benchmark evaluation will only be possible once the models are available in the public domain. The o3-mini is expected to be launched in January 2025.
OpenAI Offers Unlimited Access of Sora to Paid Subscribers
Separately, as a “day-13 bonus”, Altman announced in an X (formerly known as Twitter) post that the ChatGPT Plus and Teams subscribers will get unlimited access to Sora during the last two weeks of December. The CEO highlighted that this was possible as offices close during the Christmas period resulting in reduced server loads.
Rohan Sahai, a product lead at OpenAI added to the post and stated that the company had upgraded Sora’s blend feature and enabled shared links to let users share the AI-generated videos with friends, even if they do not have an account.
