OpenAI o3-pro benchmark performance

Hey there!

Welcome back to The Pulse, where we dive into interesting AI stories and trends backed by data, all presented through simple visuals.

> OpenAI released o3-pro - most powerful oAI model yet per benchmarks

> on par with the #159 human competitive coder

> beats new Gemini 2.5 on Artificial Analysis Intelligence Index (by 1 point), similar on GPQA & AIME '24

> o3 and o3-pro perform similarly on ARC-AGI 2

> but experts across fields emphasize o3-pro gives significantly boosted performance with detailed prompts & larger context

> o3-pro 87% cheaper than o1-pro → OpenAI cutting prices across models

> key: standard o3 prices slashed 80%, making it king of cost-performance ratio

> now cheaper than SOTA models (including Gemini 2.5 Pro) & even GPT-4o (o3 costs 20% less)

> o3 now 2-3x cheaper than Gemini 2.5 0605 to run benchmarks + sets ARC-AGI-1 frontier

> AI investments rise, but big tech acquiring fewer companies overall: 2024 acquisitions ~70% below 2020

> many acqui-hires: hiring founders + staff instead, especially in AI

> NVIDIA most active: almost 5x AI startup investments

> already made 2 acquisitions as of April this year (out of 26 total)