𝗢𝗽𝗲𝗻𝗔𝗜 𝗥𝗲𝗹𝗲𝗮𝘀𝗲𝘀 𝗢𝗽𝗲𝗻-𝗪𝗲𝗶𝗴𝗵𝘁 𝗠𝗼𝗱𝗲𝗹𝘀

OpenAI released open-weight models. Sam Altman announced gpt-oss-120b and gpt-oss-20b. This is the first time OpenAI released downloadable weights since 2019.

These models use Apache 2.0-style licenses. You have no restrictions on usage.

Model details:

  • gpt-oss-120b: 120B dense. Use this for reasoning, coding, and agents.
  • gpt-oss-20b: 20B dense. Use this for on-device or cost-sensitive tasks.

Performance data:

  • MMLU: 89.7%
  • HumanEval: 92.4%
  • SWE-bench: 63.2%
  • Context window: 256K tokens

Hardware needs:

  • Run gpt-oss-20b on an RTX 5090.
  • Run gpt-oss-120b on 4x H100s.

You can use LoRA and QLoRA for fine-tuning. Support for ollama, LM Studio, and vLLM arrives within 24 hours.

OpenAI is responding to competition from Meta, Anthropic, and DeepSeek. The open-weight market has a new leader.

Source: https://dev.to/doremonai/openais-gpt-oss-the-moment-they-finally-went-open-weight-with-120b-and-20b-models-5g4m

Optional learning community: https://t.me/GyaanSetuAi