- The Pulse by 42neurons
- Posts
- GPT 5.2 'garlic' model's benchmark performance
GPT 5.2 'garlic' model's 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.

> released just 30 days after GPT-5.1, with unprecedented >50% score on ARC-AGI 2
> 5.2 thinking deemed best for real-world professional use, performing at or above human expert level
> beats/ties experts on 70.9% (Pro = 74%) of comparisons on GDPval knowledge work tasks (making presentations, spreadsheets, etc) at >11x the speed & <1% the cost of expert humans
> 40% more expensive than GPT-5, but for huge gains
> ChatGPT “adult mode” to debut in Q1 2026
> Disney investing $1B into OpenAI + 3-yr licensing deal enabling Sora to generate videos with 200+ Disney characters

> ChatGPT now the most-installed free iOS app in the US, ~2.5 years after launch
> first hit most-downloaded status worldwide (iOS + GPlay = 46M installs) in March this year
> surpassed all social apps (Instagram, TikTok, etc.) + essentials (Gmail, Google Maps)
> in ~2 years: 0 → 2 AI apps in the top 10; ChatGPT leads, Gemini now fast-closing as rising usage + benchmark gains drive adoption & competition

> tech already AI heavy - 11x growth on top = deeper integration
> health & manufacturing still in early expansion unlike tech, although healthcare = largest vertical AI market at $1.5B spend (scribe-led)
> even slowest-growing sector at >2x growth while median sector sees 6x growth
> customer concentration highest in professional services, finance, tech (early adopters still lead scale)
> API used mainly for customer-facing apps, then customer service + content generation as non-tech firm API grows 5x YoY