I tested GPT OSS 120B on an agent project and the performance is not that much good. It doesn’t see obvious things, it repeats some contents like LLMs used to in the early days. You won’t get much benefit from this other than very low cost.
1 Like
Forget about gpt oss and add GLM 4.5, that’s a truly excellent model and equally cheap.
5 Likes
Trash model gpt-oss. It’s for babies as it has the highest rejection rate. It totally sucked for coding. Llama Maverick, GLM worked better at coding for me than gpt-oss. Qwen3-Coder is the top G in open source models for coding as of now according to the testing I did (I tested free models the entire weekend yeah coz i have no life)
1 Like