The initial reactions to OpenAI’s landmark open source gpt-oss models are highly varied and mixed
1 min read
Summary
Open AI has released two new LLMs, gpt-oss-120B and gpt-oss-20B, the first time the company has done so under an open source Apache 2.0 license since 2019.
They are the first Open AI models that can be run on private hardware and offer the potential for in-depth customisation, with no usage limits.
They have been well-received by some in the AI community, who highlight their efficiency and parity with OpenAI’s proprietary models.
However, others have criticised them for being inferior to Asian open source models and for apparently containing biases and lacking taste and common sense.
Perhaps the biggest criticism centred on the perception that OpenAI had deliberately trained the models on primarily synthetic data to avoid copyright lawsuits, meaning that, in some users’ opinion, they know “less about the world than a good 32b model”.
Overall, the response has been mixed, with some welcoming the move as a symbolic return to open models by OpenAI, while others point out that the new models are unlikely to slow down the Chinese AI giants.