OpenAI Unveils Next-Gen o3 Model Set for Early Next Year Release

OpenAI has teased its next-generation frontier model, o3, with plans for a public release early next year, following the launch of o3-mini.

OpenAI just wrapped up its 12 Days of OpenAI livestream with a bang, giving us a sneak peek at their next-gen frontier model, o3. Sam Altman, the CEO, couldn’t resist a little joke about the name, tipping his hat to Telefónica and OpenAI’s love for quirky naming conventions. Though o3 isn’t ready for your average Joe yet, researchers will get first dibs to put it through its paces for safety testing.

But wait, there’s more! Alongside o3, OpenAI teased o3-mini, set to drop by January’s end, with the big sibling o3 following close behind. How much better is o3? Well, it smashed the American Invitational Mathematics Examination with a 96.7% accuracy, leaving o1’s 83.3% in the dust. Talk about leveling up!

Mark Chen, OpenAI’s research bigwig, was practically gushing about o3’s performance. It aced standard benchmarks so hard they had to find tougher tests, like the ARC-AGI benchmark, which is basically the AI equivalent of a pop quiz on learning and intuition. Under different settings, o3 scored 75.7% and 87.5%—proof we’re inching closer to artificial general intelligence.

And let’s not forget o3-mini. This little powerhouse uses the Adaptive Thinking Time API to switch up its reasoning modes, letting you tweak how long it spends solving problems. The kicker? It matches o1’s performance but sips computational power like a fine wine, not a kid with a juice box. Expect it to hit the scene before o3 does.