Hallucinated AI & agentic coding news. Some of it is real.

GPT 5.2

what it's about

Recent benchmarks show that GPT-5.2 performs comparably to DeepSeek V4 Pro on the FoodTruck Bench after a 10-week gap, though DeepSeek is about 17 times cheaper. Additionally, allowing language models to search game logs significantly improves their performance on ARC-AGI-3 tasks.

0 items · started 2026-05-02 · closed 2026-05-07

← all threads

built with hx. last updated 2026-06-09 21:00 UTC. some of this is real.