The model is (like Composer 2) based on Kimi K2.5 and they claim SOTA performance for 1/10th of the cost. The tweet also mentions that they've started a new model from scratch on Colossus 2 (xAI/SpaceX Cluster). Really impressive how they've made this jump from being called the vscode fork with no moat just a couple of months ago.
Their previous Composer was already marketed as a cheap model capable of competing with SOTA on most tasks. The evals they shared back then backed this up but in my day-to-day usage it fell short across the board. Canceled my cursor subscription and switched to Claude Code a few weeks ago. It has its own shortcomings but in terms of model capability and UX quality Cursor will have a hard time competing in the long term. Elon Musk will be a very good way out for them.
They set themselves up for flack when they use whatever these evals are… they did the same for composer 2 which was evaled in close competition with frontier models, spoiler alert, it wasn’t even close in practice.
So now 2.5 is supposed to compete with opus 4.7? Sure…
That does not match my experience. Composer 2 was fantastic for my uses, and I hit Composer 2.5 with some very difficult things last night, which it handled fast and effectively. I don't really care about benchmarks. I care about practice, and in practice, it's been very very good for me.
Congratulations on the launch! I'm interested in trying Cursor but it's very confusing what I should buy. What does the Pro $20 plan get me in usage if I only use Composer 2.5? How fast is the model?
Ok this might be weird but I've moved everyone in my 4 person team to our team plan and costs seem to have sky rocketed compared to the individual plans. Where before most people spent 20-100 USD, now the total bill is more like 1k USD. I haven't gone into the details but it feels like I'm being scammed.
this feels super bullish on cursor/spacexai's ability to train a frontier level model. could be truly SOTA on coding given that their RL data is this powerful
Surprised this got pushed off the front page so quickly! It’s exciting to see what the Cursor team has been able to do with significantly fewer resources than the frontier labs.
I do wish they weren’t joining xAI. Something tells me there will be a contingent of researchers that departs Cursor if that merger is consummated.
"Will this decrease Revenue?" -- only if demand for high quality tokens is inelastic. If demand is instead elastic (grows with cheaper pricing) then revenue will likely increase.
"Will this lower earnings?" -- they have a current inference margin for their old models, and with the Elon deal in place, they have a new inference margin. It might be better or worse than their old one. If it's worse, then they'd need to see a concomitant increase in usage. If they don't, then yes it might lower earnings.
"Will this lower corporate value?" -- no - not least because this company is going to be owned by SpaceX approximately 90 days after IPO -- so all the new owner will care about is being benchmark competitive with Anthropic and oAI for the first n quarters. If they can do that, it will massively increase the corporate value of SX; it's hard to build a frontier lab.
It's a bit confusing to me why they'd make this 'fast' version the default, as it appears to be much more expensive than Composer 2. Wasn't it supposed to be a very cheap alternative to SOTA models?
I kind of want to try it, to see if and how far they can take an open model and improve it but I really don’t miss the Cursor user experience. Constant UI changes, half-baked features, smaller and smaller limits, useless AI change attribution; I think I’ll wait for others to report if it’s any good.
Noticed recently they keep opening their “Agents” window when the project was last opened in the VSCode fork window in the hopes I’ll just continue working in that when the UI is totally different and missing things I need.
For a professional tool it’s getting egregious how little respect they have for my workflows and flow state they way they keep moving, changing iconography and flipping switches of the UI.
It’s clearly being ran by someone who comes from a social app or sales app growth hacking background.
> It’s clearly being ran by someone who comes from a social app or sales app growth hacking background.
I fixed that by using cursor the agent but not the UI.
I'm just running cursor in GNU Emacs via agent-shell (https://github.com/xenodium/agent-shell). Their cli client (aptly named "agent") supports ACP (agent client protocol) so the UI can be skipped altogether.
I know this sounds like a meme ("use x in emacs") but at this point at the very least i can keep my workflows and my UI all the same and focus on my work rather than "where did $company put $feature this month".
I gave up, canceled my plan, and went back to boring old VSCode. It feels so much more stable, and my Mac no longer runs out of memory. With cursor I had to reboot my macbook several times a week and had to always be plugged in.
I agree. I quit cursor and replaced it with conductor and a mix of Claude Code / Codex/ Copilot and i dont miss it as such. Maybe one day I will come back.
I try it from time to time and feel the same way. Some people I know really like it but I can’t tell if that’s because it’s good or just because it’s what they’ve become familiar with and they don’t like to change tools. Cursor had a good head start and a lot of early PR.
Sounds like it's the last Kimi-line model at Cursor? As expected they say they'll be training a larger model on the SpaceX infrastructure, or have already started most likely.
I'm very curious to read about the Composer 3 architecture when it comes out. More frontier coding models are a good thing, especially if they diversify into different strengths/weaknesses.
65 comments
[ 2.7 ms ] story [ 64.3 ms ] threadSo now 2.5 is supposed to compete with opus 4.7? Sure…
I do wish they weren’t joining xAI. Something tells me there will be a contingent of researchers that departs Cursor if that merger is consummated.
Wouldn't this compress ai revenue like 15x quickly
If they really have a 4.7 opus high equivalent at 1/16 the cost wouldn't this significantly effect all the current capex and planing
Maybe they are getting elon to cover cost
"Will this decrease Revenue?" -- only if demand for high quality tokens is inelastic. If demand is instead elastic (grows with cheaper pricing) then revenue will likely increase.
"Will this lower earnings?" -- they have a current inference margin for their old models, and with the Elon deal in place, they have a new inference margin. It might be better or worse than their old one. If it's worse, then they'd need to see a concomitant increase in usage. If they don't, then yes it might lower earnings.
"Will this lower corporate value?" -- no - not least because this company is going to be owned by SpaceX approximately 90 days after IPO -- so all the new owner will care about is being benchmark competitive with Anthropic and oAI for the first n quarters. If they can do that, it will massively increase the corporate value of SX; it's hard to build a frontier lab.
Should try their CLI!
For a professional tool it’s getting egregious how little respect they have for my workflows and flow state they way they keep moving, changing iconography and flipping switches of the UI.
It’s clearly being ran by someone who comes from a social app or sales app growth hacking background.
I fixed that by using cursor the agent but not the UI.
I'm just running cursor in GNU Emacs via agent-shell (https://github.com/xenodium/agent-shell). Their cli client (aptly named "agent") supports ACP (agent client protocol) so the UI can be skipped altogether.
I know this sounds like a meme ("use x in emacs") but at this point at the very least i can keep my workflows and my UI all the same and focus on my work rather than "where did $company put $feature this month".
I gave up, canceled my plan, and went back to boring old VSCode. It feels so much more stable, and my Mac no longer runs out of memory. With cursor I had to reboot my macbook several times a week and had to always be plugged in.
https://cursor.com/docs/cli/installation
https://github.com/xenodium/agent-shell
Really nice to see they're giving credit to the company and I am optimistic Kimi K open models soon will outperform Opus models
I'm very curious to read about the Composer 3 architecture when it comes out. More frontier coding models are a good thing, especially if they diversify into different strengths/weaknesses.
Hard to outperform the model you distill...