▲ 171 points
back
104 comments
I'm a bit skeptical.
Cursor's benchmark finds that Cursor's model (Composer 2.5) is basically as good as Opus 4.8 max and GPT-5.5 xhigh, but at a fraction of the price.
Artificial Analysis' testing shows Composer 2.5 to be pretty far behind: https://artificialanalysis.ai/agents/coding-agents. You look at the DeepSWE benchmark (which is probably the hardest to game at this point) and GPT-5.5 xhigh gets a 64, Opus 4.8 max gets 56, and Cursor 2.5 gets 16.
I don't doubt that Cursor works well for some people. It's beating DeepSeek v4 Pro in the DeepSWE benchmark and that's a very capable model. But I'm skeptical of the claims that it's a competitor for Opus 4.8 and GPT-5.5. It just seems convenient that their model does so well on their own benchmark while third party benchmarks have it far behind. Maybe it's a really great benchmark and a better measure than third party ones - I'd love for a cheap model to do as well as the expensive ones.
(I work at Cursor) When Composer 2.5 launched, we initially scored very competitively on AA's composite benchmark. I believe 3rd place overall. They have recently updated to use DeepSWE, which has more of a focus on very long-horizon tasks, and Composer isn't as good at those yet. We're aware and working on this for our next model.
Overall, some benchmarks show Composer doing well, others not so much. We think the model is very capable at the given price point. There's lots to improve! If you see any specific behaviors or places the model isn't very good, lmk here or can email me lrobinson at cursor.com.
Not hard to understand what's going on here. They RL'd around patterns in their data and specific capabilities, so of course they'd construct a benchmark that's aligned with the training set.
Ironically, their benchmark might be more accurate than artificial analysis for a narrow slice of things that Cursor's Eigencustomer is really interested in. Otherwise I'd take it as just another data point.
DeepSWE is slightly flawed in the sense that is uses only its own harness and that causes issues on models that are not correctly supported by it. There's huge amount of evidence that the harness plays a big role in how these models work and yet DeepSWE entirely removes that (and has probably only tested that it works fine with some favourite model of them).
There's also issues with cost calculation (as that harness doesn't use caches) and so on as reported on their github issues.
None of the benchmarks are perfect, but that does explain a lot of the variations between benchmarks.
Cursor sessions are pretty much what composer models are RL'd on. This bench and the training data are/should be basically the same distribution.
Anecdotally, I find Composer 2.5 to be useless. I do use light LLMs like Claude Haiku and some of Cursor's older free models, but Composer is negative productivity for me.
Naturally, given it’s their benchmark they have overfitted their model somewhat to it.
For lighter interactive agentic coding, where you type stuff into an IDE and a minute or three later get results back for review, composer 2.5 is honestly pretty great. The results get notably worse for larger tasks though.
I can't speak to benchmarks, but I have used Composer 2.5 extensively and it's performed quite well in my real world tasks.
that benchmark seems to match my experience. GPT 5.5 is significantly better than Opus 4.8, last time I tried composer 2.5 it was truly dumb, and Fable to me looks to be on par with GPT 5.5 but .. different overall ... The best is to have a LLM-peer-review between GPT and Opus (now Fable) for best outcome.
By the same token, Fable 5 is given a score of 77 vs 76 for GPT 5.5
> Cursor's benchmark finds that Cursor's model (Composer 2.5) is basically as good as Opus 4.8 max and GPT-5.5 xhigh, but at a fraction of the price.
Your skepticism is well-founded IMHO. I have found that if you are one-shotting a Django/Next CRUD app, a typical React/Vue UI, shell scripts or GitHub Actions, Composer 2.5 is fantastic!
But for anything outside the median of the last decade's web development - like free-body physics, kinematics, or optimization - Composer is horribly unpredictable.
That's what makes it _dangerous_ IMHO.
It isn't universally trash! Rather, it confidently makes subtle, incorrect assumptions. It will hallucinate formulas that don't appear in your specification and design docs. Then write tests that pass it.
It inserts tiny footguns that require you to scrutinize every single token it generates. At that point, I would rather be coding by hand.
Opus 4.8 max, on the other hand, refuses to guess, atleast the way I have set it up. If there's any ambiguity about the implementation or how tests should be written, it stops and asks me for clarification. I actually trust the output without worrying about hidden disasters and ticking timebombs. I can confidently review the test suite, add a few edge cases on my own, spot check the code and be comfortable knowing there are no disastrous footguns lurking in the shadows only to come out in the darkness of production deployments.
Let me repeat - Opus 4.8 max stops and asks me for clarification. It writes the tests I would have written. It writes tests that fail, exposing gaps and errors, that then allows me to iterate.
Composer 2.5 OTOH will run with whatever it decides I meant and write something that steals productivity, not add to it.
Same harness (Cursor), same rules, same prompts, vastly different outcomes!
Yes, Opus is far more expensive, but it's worth it for the time saved on review and refactors, which are our current blockers.
The real friction is that Cursor's marketing is so aggressive that the people paying the bills look at my Opus usage and demand to know why I'm not using the cheaper alternative!
It's an impossible argument to win when the rest of the company's devs are happily building standard web apps on Composer without issue, blissfully unaware of how the model not only falls apart but is just unreliable on harder engineering problems.
Fable 5 is on a league on its own. If history in the LLM space is any predictor of the future, in ~6 months (Q1 2027) we should have open weight models that are competitive with Fable 5. Without considering what it will take to run such a thing, I would be extremely excited to have open access to such a capability. Great times ahead!
I mean, they train their model on their training data. So by it should score well on their own benchmark.
Composer writes the worst, stupidest, most naive and straight up brains-dead code you could imagine. Fast and cheap is about all it’s got going for it. I mostly use it for “sort these lines alphabetically” and stuff that’s a smidge too complex for regex find/replace.
I'm pretty baffled by their choice of axes. I would have thought that the left was the cheapest, not the most expensive. I appreciate that this layout means that top right can be best, but it's still unintuitive to have this backwards cost axis IMO.
Putting that aside, I spend all day every day implementing very, very hard things right on the edge of what agents are (barely, sometimes) capable of, and I have had to keep Opus on max for things that need 'real validation' for a while now. And that has felt like 'the only way' to get Opus to perform even close to 5.5 xhigh. I'm only using Opus at all because GPT-5.5 in the subscriptions only has a small (400k, but 258k effective) context window.
The difference is that 5.5 xhigh is extremely fast in most practical cases, both efficiently implementing _overall_, and responding very quickly with great adaptive thinking if you ask it something that it doesn't have to think about. Opus 4.8 Max will needlessly chew on everything and can take hours to implement even simple things, so I can mostly only use it for planning/review.
Fable is much much better at adaptive thinking / responding quickly (although probably still worse than 5.5 xhigh), and... I think folks have said enough elsewhere about its strengths and weaknesses. Sadly still not a reliable implementor for my hard tasks though (that's still GPT's domain) – it tends to leave big, dangerous holes hiding inside implementations unless babied.
>Putting that aside, I spend all day every day implementing very, very hard things right on the edge of what agents are (barely, sometimes) capable of
Is a single thing in your post demonstrable, or are we just supposed to take your word for it? Because all of this stuff sounds laughably subjective.
>it tends to leave big, dangerous holes hiding inside implementations unless babied.
A brainwave: perhaps GLM or DeepSeek could be integrated into the mix for the purposes of red-teaming the code. Fable has been blinded to security by design[0], and the open models are pretty decent at it.
[0] It's not clear what the situation with GPT-5.6 will be but the blog suggests similarly over-cautious safety filters.
Amusingly the posts for recent Opus releases brag that they successfully made it worse at security! "during its [Opus 4.7] training we experimented with efforts to differentially reduce these ["cyber"] capabilities"
It’s Gartner. Top-right is where you want to be.
> it tends to leave big, dangerous holes hiding inside implementations unless babied
it's fascinating that I used these same exact words to express my distaste for Composer and my preference for Opus. I suspect, the domains and problems we are trying to solve need to be shared. I wrote about it here: https://news.ycombinator.com/item?id=48766275
Would love to reach out to discuss more, if you're ok with it, or absolutely feel free to do the same as my email's in the profile like yours!
> I'm only using Opus at all because GPT-5.5 in the subscriptions only has a small (400k, but 258k effective) context window.
Do you find that makes a difference in your work? I've been using 5.5 high/xhigh to optimize and benchmark a C codebase, and just reading the initial code virtually fills the first context window. A session will auto-compact 5-15 times, but it seems to do okay in spite of that because the task is mainly focused on the latest window each time.
I think for programming the strength of GPT over Opus is winning here over the context window.
I agree why they reverse the x axis makes this graph very hard to understand for the casual observer.
You can set GPT 5.5 to 1M context mode in Cursor but it costs more after the default 272k.
> I'm pretty baffled by their choice of axes
To put their own model out in front?
opus@max is on average worst than opux@xhigh
for supporting evidence, see first chart here: https://www.anthropic.com/news/claude-fable-5-mythos-5
I mean, they basically did that to make sure that Composer 2.5 sits on the right side, and to a quick reader looks good
It's hard to believe Composer 2.5 is that good. I tried to compare it with GLM 5.2 or Opus 4.6 and it lacked thinking about the problem and critical reasoning. It's great for executing plans made by other models, but even then it does some weird code manipulation that is far from how other files around actually work.
I'm not using Cursor at the moment, but when I did (not too long ago) my experience was similar. Plan with Opus, implement with Composer, clean up with Opus.
Composer did a competent but not amazing job with a good plan. What I really liked though is it was fast! Opus could take 30 minutes to do something Composer would get done in 5-10 minutes. Of course the output wasn't perfect, but that's why I'd do a cleanup pass using Opus or Codex.
It's all a balance though, constantly changing and completely dependent on the problem you're solving. I just remain flexible and adapt my process to what's working best in the moment.
Interesting that Opus 4.7 does better than 4.8. Too bad they didn't test 4.6, too. I witnessed a man here mocked yesterday for insisting it was better than its successors!
Although, the benchies are always tricksy ... On DeepSWE, GPT-5.5 beats Opus-4.8, by a fair margin, but on FrontierCode, the situation is the other way around.
The only benchmark you can trust is your actual workload!
everytime a new benchmark appears, Chinese models are far lower than the level where they are supposed to be according to existing benchmarks. then after a while they recover :)
Composer 2.5 is really effective at some tasks, but doesn't do as well on higher complexity tasks from my own experience.
With that being said, Cursor Bench is nonetheless a fantastic gauge of LLM quality and performance. The most interesting outcome of Cursor Bench is Fable 5 to GPT 5.5, its almost a perfect continuous line. As of the report, Fable to GPT might be the new standard of agentic programming/building
I did my own rounds of tests for a lot of models.
I made multiple rounds of comparisons for issues in my open source project using Fable 5, GPT 5.5, Opus 4.8, and Composer 2.5
Made them work on various complexity issues
Here are the reports, i recommend using a Sonnet 5 or better model to summarize them because its quite a bit of information to consume across all the various tests: https://www.richkuo.com/#llm-battles
My takeaway from Composer 2.5 is that its better to use Opus or GPT for planning, and then Composer to build, use Opus/GPT to pr review, and have Composer fix findings, and loop that process. Incredibly token efficient, fast, and gets the same quality as if you worked the whole thing with Fable/Opus/GPT.
I wish all these sites would show pareto frontier graphs of cost/performance. That's the main 2 things that matter (I guess you could make it 3D with a speed param as well). https://paraplouis.github.io/llm-pareto-frontier/ is the best of these graphs I've seen but it doesn't update as frequently as I'd like.
If I understand the graph correctly;
Fable is using less tokens to achive that same tasks compared to sonet and opus. If so that is a good thing. It feels like we for a while there was spitting out tokens to get a better result. If the model themselves are getting better without generating more tokens that feels like a real win.
Q1: Why is number of steps relevant in this graph? What does it tell us?
Q2: and why have they flipped the horizontal graph so that 0 is to the right and not at origo? Is that some kind of new smart thing? can't say i have seen it before
Cursor’s model excels at Cursor’s benchmark; news at 11.
The other models however are reasonably where I’d expect them to be from experience piloting all of them. Fable is outclassing everything at most things at 10x the cost, but sometimes it isn’t a choice between cheap and expensive, but expensive and possible; I’ll need to learn where that boundary is just as it was the case with other models.
I've used both Composer 2.5 and GPT 5.5 (both in Cursor and in Codex) extensively, and their claim that Composer 2.5 is anywhere close in performance to GPT 5.5 is absolutely farcical. It's faster, but it's nowhere near as good.
And given that you can only use Composer with a Cursor monthly subscription, cost comparisons are pointless since an equivalently priced OpenAI subscription gets you just as much usage of the better model.
The most interesting part is costs . Gpt 5.5 and sonnet 5 cost same amount of money as GLM 5.2 but are more capable models
Very skeptical about the composer accuracy. I have been using it for 6 months now and it is very fast, especially compared to anthropic models, but the result it produces, especially with more difficult tasks is very shallow. It feels like it just finds the cheapest way to deliver the task.
backwards X axis? is there a reason for that? it looks ridiculous
I like Composer a lot as a general-purpose workhorse, but putting it over gpt5.5 medium makes the whole graph lose trust to me, asme witg GLM so low
I feel like this benchmark reiterates my disbelief that anyone uses the latest Anthropic models for any productive work. They seem to be the best at burning tokens and spawning unnecessary subagents even for well-defined and tightly scoped tasks.
Can we get a count of people that have had Claude read irrelevant documents or perform unnecessary web searches even when told not to from the beginning?
I'm starting to wonder if this increased token usage is inadvertently bleeding into how Anthropic actually trains their model, especially leading up to IPO. As older models are deprecated and users are forced onto newer models, if the default is less efficient and more token expensive that directly results in higher "profit" for Anthropic in terms of the consumption their users have to tolerate - lest they jump to a competitor.
No shot 2.5 is beating out 4.8
Do these benchmarks even add any value at this point? This one is basically Cursor saying that their model is as good as the frontier ones at a fraction of the price. The independent benchmarks are probably part of training data now and the models are pattern-matching against them all the time. The final test of a model (and the harness, probably) is how good it works FOR YOU - since most of the models can pretty much do most of our tasks on a daily basis - it boils down to which one has the least friction to its usage.
AI is getting expensive.
Why would anyone take this benchmark seriously? Cursor is obviously biased here. They can design it and its presentation however they want to tell the story they want to tell.
Would like to see wall times. I feel that’s the part that annoys me most, my tasks aren’t particularly challenging I want them done fast
is composer 2.5 that good at that pricepoint? Seems like the gemini flash playbook of trying to get most bang for the buck.
Cursor: Find me another benchmark where Composer 2.5 is a top 10 frontier coding model
insert obama medal meme