Feature request for product/service
AI Models
Describe the request
Now I know Composer 3 has not been officially announced or confirmed, so that limits how much any cursor employee can speak on it but I just wanted to throw this out there
Given that Qwen 3.8 27B is matching GPT Luna and Opus 4.6 (and Composer 2.5) in many artificial analysis benchmark suites - a good majority of work is becoming viable on home GPUs. Ive been blown away by having Grok 4.6 plan and then having my local model execute that plan as it really takes away a lot of the hang up points it can get stuck and burn the limited context on and it really shines when a lot of the “figuring out” has already been done and it can focus on execution
As such I think one thing that could benefit both users and cursor/spaceX AI is if/when Composer 3 comes out, Cursor entered the small weight arms race with a Composer 3 27-31B range distillation for home workstation users. I think it carries huge mutual benefits, not the least of which would be giving something to build the local arm of cursor around which is sorely lacking after they bought continue but dont seem to be doing anything with it in terms of integrating local LLM support. My continue extension is very much jerry rigged with AI edits at this point to be modern.
It could differentiate further Composer and Grok and really give cursor another massive W in its recent streak of Ws and winning over the entire AI ecosystem. I think Qwen 3.8 27B proves that Home LLMs are here to stay and are going to be very viable going forward and the first ecosystem/harness dev that realizes this and incoporates it into the workflow of their software is going to be very well positioned. Heck it could even become Cursor’s 3rd “Leg” like how GPT has Sol, Terra, Luna and Anthropic has Fable, Opus, Sonnet. Imagine if you will a system so fluid that a user could either user composer cloud on low as “Luna” or use their home LLM running composer 3 27B in a Luna/Sonnet role and its all built in. Heck Cursor could even go so far as to integrate a user’s LLM and Composer 3 27B into the auto workflow automatically if present so the auto model intelligently selects when the home LLM is right for execution.
Imagine if you will a setup of
Composer 3 27B locally - The majority of basic execution
Composer 3 cloud- The majority of higher code writing, bug fixing and greenfield code writing
Grok- Planner, executes the hardest tasks, codes, refactors and debugs, starts greenfield code scaffolding
Cursor seems to pride itself on being the king of integration and home LLMs have been a massive gap that has only grown bigger as they have recently become far more viable. Composer 3 27B open weights + integration would be a chance to close that gap and be the model of choice for home LLM users.
Thank you for coming to my TED talk on this matter.
Operating System (if it applies)
Windows 10/11