Hi! when will OpenAI: GPT OSS 120B model be added to the available model list? It’s pretty good in agentic coding (if it really is horizon beta). Also, it’s very cheap to run. Will it come as a free model or as auto?
8 Likes
Any benchmarks? I doubt it is horizon actually, they say oss models are almost as good as o3, not better
I mean, if it’s so easy to run, maybe it could be the new “auto” by default? Or actually it’s too expensive or not so good?
well Sam Altman just said that it’s at the level of o4-mini
So, it’s not the best model out there, but I think it’s the best open-weight reasoning model which is basically cheap to run so it just gives us a cheap yet good option in Cursor.
You are confused btw, this is not Horizon. Horizon is a separate model by an unknown provider, it’s still under testing.
The GPT OSS is the first open weight reasoning model from OpenAI
A combo of GPT-OSS 120B + Qwen 3 Coder + GLM 4.5 + Kim K2 is like the dream scenario for AUTO mode.
5 Likes
OpenAI just released a banger oss model:
https://openai.com/index/introducing-gpt-oss/
Can we use it with Cursor plans ASAP?
Up on fireworks already: OpenAI GPT OSS 120b API & Playground | Fireworks AI for $0.15/$0.60
TIA Cursor Team 
well, horizon beta says it’s created by openai… I was using it with kilocode and it said that. Unless you have some counterproof, it’s most likely horizon beta. Also, it’s fast as a 120b model is supposed to be
Interesting, it said it was made by Anthropic to me 
1 Like
I have a feeling it might be opus 4.1
with that speed? It must be running in a quantum computer then
1 Like
Honestly the horizon’s are really good at React UI, that’s the only reason I suspect this something by Anthropic. The speed is so good too
1 Like
Bro, GPT OSS is a reasoning model
Horizon Beta is not a reasoning model (or at least the reasoning is disabled on openrouter)
A qwen model used to say it’s created by openai because it was trained on openai’s data.
claude sonnet 4 says it’s sonnet 3.5
so, this all doesn’t prove anything
why would you see horizon beta still up on openrouter ?
Horizon beta was suspected as GPT-5 but watch Theo’s video and tweets, he thinks it’s not GPT-5 and it does seem to make sense because from Sam Altman’s tweets it appears GPT-5 will be released later this week while Horizon Beta just came 5 days ago for testing and will probably stay for a while before it can be released.
Edit: This reply was meant for @Corvo_Prophet
2 Likes
it’s above almost everything else for coding I ever used. I think it loses only to anthropic sonnet 4 and opus 4. But it’s so fast and cheap that reprompting or retrying a prompt feels worth still. Well… about horizon beta being still up, maybe it’s because they want to keep training on data and improving it. And I don’t have an answer to the reasoning part…
1 Like
Forget about free or base, just add them at all. Cursor already uses fireworks and they have the best pricing right now for the gpt-oss models.
Cursor was quick to add Claude 4.1 (which should not be used unless you want to fill your entire pro subscription in a day) but not gpt-oss.
Let’s get these new models on Cursor asap please!
2 Likes
Please guys let’s try to focus on glm 4.5 keep telling the team to add each new model to the person if in the place of the auto is to defocus their attention
6 Likes
I got the new Open Source OpenAI model 20b running locally, but it doesn’t work great yet. Its giving me markdown in the response and just generally not working like the agent does.
Some optimization specifically for this model will be huge!
1 Like
please add, maybe qwen coder as well…
1 Like
Andrej Karpathy has touch on this, if LLMs are not finetuned or instructed about who made them - they usually default to OpenAI because of the training data so I wouldn’t take that output at fact. But, can’t tell for sure!