AI models’ performance on Aider Polyglot coding benchmark

Hey there!

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

> Gemini 2.5 Pro cheaper in API pricing, but o3 more token-efficient leading to relatively lower price

> 32k thinking across models gives significant performance boost at slightly higher costs

> Gemini models becoming costlier over time while OpenAI lowers price, slowly leading to convergence

> models currently more cost-competitive compared to few months ago

> o3, o3 (high) & DeepSeek R1 show best value

> Claude still developer-favorite despite performance on benchmark

> expected 56% fall YoY in operating profits

> from ₩10.44T in Q2 2024 to ₩4.6T Q2 2025 now vs expected ₩6.26T

> revenue to hit ₩74T (~$54B)

> fall due to lack of HBM sales to NVIDIA against competitors SK Hynix & Micron + US restrictions & tariffs

> to have <₩1T in profits from semiconductor division vs last quarter’s ₩1.1T (1/7th of SK Hynix)

> xAI replaces Anthropic in 2nd place after its acquisition of X

> also most funded after OpenAI ($17.4B) with 500x revenue multiple, compared to OpenAI’s 33x & Anthropic’s 31x

> American companies valued highest collectively, followed by China then France (due to Mistral)