I think Grok is too, at least an older version. There’s also gpt-oss, and Meta has released a lot of “open source” models, but I think they use weird licences. Meta and Deepseek (and Alibaba) researchers publish papers that are actually useful, while the rest just publish marketing material, trying to keep the research itself private.
Gpt oss is borderline crap, it’s not that smart, not that great and it’s pretty censored, but it can have niche uses for programming. The oss 20b in particular can be easier to run in some setups than their competitors like Qwen 3-30b. oss 120b is quite heavy: the cost to performance ratio is not good.
Meta abandoned the open source ideal since Llama 4; they went closed source.
Older open source versions of Grok are literally useless, no one should use them. Their cloud closed source models are decent.
I think Grok is too, at least an older version. There’s also gpt-oss, and Meta has released a lot of “open source” models, but I think they use weird licences. Meta and Deepseek (and Alibaba) researchers publish papers that are actually useful, while the rest just publish marketing material, trying to keep the research itself private.
Gpt oss is borderline crap, it’s not that smart, not that great and it’s pretty censored, but it can have niche uses for programming. The oss 20b in particular can be easier to run in some setups than their competitors like Qwen 3-30b. oss 120b is quite heavy: the cost to performance ratio is not good.
Meta abandoned the open source ideal since Llama 4; they went closed source.
Older open source versions of Grok are literally useless, no one should use them. Their cloud closed source models are decent.
Deepseek and Alibaba’s models like Qwen are good.