Gemini 4 Neon tops a real-repo bug-fixing test by a hair
Paweł Huryn's test put Neon at 63 against GPT-6.2 Astra's 62.7.

Paweł Huryn said he tested Gemini 4 Neon on real work: two repos with 105 planted bugs, scored on how many it could find and fix.
His results put Gemini 4 Neon (xhigh) at 63, GPT-6.2 Astra (max) at 62.7, GPT-6.2 Astra (xhigh) at 60.2, Fable 5.2 (max) at 58 and Gemini 4 Carbon (high) at 57.3. "It's the new #1," he wrote.
The post passed 876 likes Thursday morning.

