Estebankiwi@estebankiwi

フォロー

AI coding benchmark comparison

Local AI models are now surpassing GPT-5 on coding benchmarks. Consider this: Gemma 4 31B achieves a score of 38.7, Qwen 3.6 27B scores 36.5, and GPT-5 trails at 36.0. These are models you can run on a MacBook, outperforming what was considered the smartest model in the world just nine months ago. GPT-5.5 currently leads at 59.1, with Claude Opus 4.7 at 52.5. The gap between frontier models and local ones is still significant, but it is closing every day. It seems inevitable that we will soon have models rivaling GPT-5.5 and Claude Opus 4.7 running on consumer hardware. The trajectory is unmistakable.

3mo

Estebankiwi@estebankiwi

フォロー

AI coding benchmark comparison

Local AI models are now surpassing GPT-5 on coding benchmarks. Consider this: Gemma 4 31B achieves a score of 38.7, Qwen 3.6 27B scores 36.5, and GPT-5 trails at 36.0. These are models you can run on a MacBook, outperforming what was considered the smartest model in the world just nine months ago. GPT-5.5 currently leads at 59.1, with Claude Opus 4.7 at 52.5. The gap between frontier models and local ones is still significant, but it is closing every day. It seems inevitable that we will soon have models rivaling GPT-5.5 and Claude Opus 4.7 running on consumer hardware. The trajectory is unmistakable.

3mo