مدل بزرگتر رو یه کارت NVIDIA اجرا میکنه، مدل کوچیکتر رو میتونی روی لپتاپی با رم ۱۶ گیگ استفاده کنی.
📊 عملکردشون تقریبا با مدلهای o‑series برابره و از مدلهای باز دیگه بهترن، ولی هنوز کمی ضعیفتر از o3 یا o4‑mini هستن.
OpenAI just launched two new open‑weight reasoning models called gpt‑oss‑120b and gpt‑oss‑20b. They’re free to download on Hugging Face. The bigger one runs on a single NVIDIA GPU; the smaller one works on a regular laptop. They perform about as well as OpenAI’s o‑series models, and better than other open models—but still a bit weaker than o3 or o4‑mini. The catch is they hallucinate more. These are the first fully open models since GPT‑2, under an Apache 2.0 license.
📰 @scopeofai | #news