Rendered at 15:17:17 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
robbomacrae 8 hours ago [-]
SCB is definitely an underrated benchmark. For me the unique selling point is that it more closely mirrors software development by not stopping after a single task. The agent has to keep code clean. The only disadvantage is all the problems are greenfield and not git inited so the agents don’t make use of git diffs.
There is a small but growing community on discord for discussing SCB so if interested please join https://discord.gg/BrC4BA9sVj
dhorthy 22 minutes ago [-]
so wait is the finding that most of those skills reduce pass rates against SCB? wild
jrflo 45 minutes ago [-]
The author of the original paper did not include error bars on the cost and quality results, which is a bit alarming. They did include a +/-, but did not list explicitly if that means standard deviation, range, or 95% CI, etc. Another table is listed as that being one standard deviation. If we take those results to be a standard deviation as well, I think we can conclude that there is no statistical difference between the outcomes of all models in terms of quality and cost, as the error bars are all overlapping.
WilcoKruijer 7 hours ago [-]
I hope the big labs will start using this benchmark in their RL pipelines. Reducing complexity in generated code should be the number 1 priority, in my opinion. The holy grail for me is models implementing features while reducing LoC (i.e., choosing the right abstractions).
What is also nice about this benchmark is that it can be used to iterate on prompts/skills for reducing code complexity.
falsemyrmidon 5 hours ago [-]
It's just as easy to go too far in the direction of shoving too much logic on a single line in the name of reducing lines of code.
Normal_gaussian 9 minutes ago [-]
It should be hoped that whoever is doing this is using automatic formatters; though of course metrics relating to code complexity are best done as measures over the AST.
blcknight 5 hours ago [-]
> I hope the big labs will start using this benchmark in their RL pipelines.
Labs do not train on benchmark data (allegedly). They can train on similar problems, but benchmarks have specific strings in them that labs are supposed to be aggressive in filtering out of their training corpora.
jerf 1 hours ago [-]
I would at least consider the possibility that they already are, and this is how it is going. It could be a fundamental architecture problem for LLMs. It's not like "slop" is a new problem and I'm sure they'd love to announce a new model that generates much less "slop". The fact they've never so much as mentioned it suggests to me that it's not something they've been able to fix.
dmrivers 3 hours ago [-]
I always have Claude recite a pledge before starting coding to fix redundant code it notices over time. It does seem to find redundancies, but only when I point out bugs, that's when it goes into fixing mode and actually applies my Don't Repeat Yourself preference from the CLAUDE.md.
The original paper cited by this post does try to see if improved prompting will make a big difference in the end using a `plan_first` prompt variant, but find no influence on pass rate at the end of the benchmark. The `plan_first` seems to assume coding agents will just refactor once they finish features, but I don't think they tend to refactor significantly unless they are told to fix bugs rather than build features. The benchmark leaves tests hidden with no fail-to-pass feedback, so that may be why degradation is monotonic.
try-working 3 hours ago [-]
This is just superstition.
notnaut 3 hours ago [-]
“Coding” can be more like communing with an otherworldly presence via esoteric gesturing nowadays than ever. Superstition leads countries and companies.
KyleTheDev 2 hours ago [-]
Every day we get closer to being mystics having to commune with orbs in order to create phenomena.
2close4comfort 53 minutes ago [-]
yeah like the palantíri in Lord of the...oh god its too late.
organsnyder 1 hours ago [-]
Musk definitely gives Dukat vibes.
huflungdung 2 hours ago [-]
[dead]
Vgoose 15 hours ago [-]
Nice! I actually ran across this paper+benchmark recently, too. It's the first I've found that start to aim at some of the non-functional and longitudinal requirements that I think have always been an important part of writing production code.
It's especially relevant now that models are good enough to solve ~most point-in-time problems.
Some relevant but disconnected thoughts:
- deterministic scores are so nice
- what "maintainable" is is probably some high dimensional space described by these signals; it'd probably require some human labeling to figure out where this space is
- another signal I've been thinking about and I'm seeing increasingly get brought up is the state space of a system; I'm seeing formal methods pop up a lot recently
NitpickLawyer 11 hours ago [-]
> another signal I've been thinking about and I'm seeing increasingly get brought up is the state space of a system
State space of a system AND the way to make it accessible / visible to a model. Many times a model can work magic if it can "see" the state of a system in a way that suits it. That's why sometimes having a cli added to the environment seems like such a big unlock. Because that cli usually takes a complex state and allows visibility into it, and possible manipulation in a structured way.
Vgoose 1 hours ago [-]
Ive have a definition of state space that is calculated by just the types of the system. It’s a rough approximation of the true state space but it’s convenient (and more reflective of what we actually mean by system state IMO) and types are inherently accessible to the model. I recently put down some thoughts around working with types to improve communication w/ AI that kind of sets this framing up: https://www.alecvo.org/blog/types-with-ai/. The actual definition is still WIP.
nicoty 8 hours ago [-]
I've been thinking a lot about this recently. In my case: how do I get an agent to see the important parts of the current plan and get it to stick to it without deviating, especially as loops get into longer and longer cycles and compactions erase prior context?
I think the problem with just encoding a whole plan in a single markdown file is that it gets polluted really quickly (agents can stop adhering to instructions to keep it clean and conform to a specific structure), which makes it harder for agents to see which parts of the plan they should focus on. As such, I've reached a similar conclusion about giving the agent access to CLI tools to help them deal with this. To try to mitigate this, I've been getting Claude to develop for me a CLI tool that:
1. Scaffolds reusable, structured plan templates and a reusable workflow that structures how to tackle the plan, step by step.
2. Validates that the plan files still conform to the correct, parseable structure.
3. Parses and evaluates the plan files, to determine what the current state is and what the next valid transitions and states are according to the workflow, like a state machine, and outputs instructions and reminders for agents as to what they should do at each step of the workflow.
So far, I've been dogfooding the tool and it seems promising: I can leave Claude running for longer and it doesn't drift as much. However I haven't ran any benchmarks yet and I'm still not entirely happy with the state of the codebase ( https://github.com/nothingnesses/agent-scaffold ), so take this with a grain of salt.
That said, I'm also bullish on using agents with formal methods and proofs. Type checkers and compile-time checking in general are great because they surface errors early and with great specificity. So if you can encode your specifications with, e.g. dependent types, you can use the type-checker as a way to steer the agent when it goes wrong and gets off-track.
dhorthy 14 hours ago [-]
> - what "maintainable" is is probably some high dimensional space described by these signals; it'd probably require some human labeling to figure out where this space is
this is a nicely succinct way to put this - a multi-dimensional space where no single metric is really useful
state space of the system is interesting too. I would guess that for any production software with dependencies like databases/third parties that might be too hard to measure, but if you can silo off parts of your system into bounded state machines, it may be a value metric on some module behind a clean interface.
I think the kubernetes control loop model is a great instance of this, a handful of scoped components that own a control loop across a well-defined state machine, that can operate / recover in the face of most network partitions or downtime - the promise of CRDTs but rather more a pragmatic approach to it
sothatsit 14 hours ago [-]
This matches my experience of Opus 5 being a nice improvement over Opus 4.8, but not being revolutionary like Fable felt.
I’ve now replaced my use of Opus 4.8 xhigh with Opus 5 medium, and I’m using less tokens and it’s quicker. I can understand people being annoyed by its writing style but for getting work done that really doesn’t bother me. I’ve been really enjoying using it.
rubicon33 14 hours ago [-]
I think they neutered Fable. When it first came out it was indeed revolutionary. But what we have today, is not what we had before the ban.
swader999 14 hours ago [-]
Noticed that too. I wonder if these things just degrade over time, perhaps with the way it writes memories about my project as it goes
Espressosaurus 14 hours ago [-]
I’ve observed the degradation, but I suspect what’s happening is they’re tuning it for lower inference costs. Maybe turning down the amount of thinking, maybe quantizing, maybe something else.
It seems like there’s a week by week and sometimes day by day change in performance when on a subscription plan using their harnesses.
conception 13 hours ago [-]
https://marginlab.ai/trackers/claude-code/ their tracker generally shows that isn’t the case. The only times I’ve seen it drop is something broken and just before fable launched.
nerdsniper 10 hours ago [-]
I mean they could just be routing known benchmark questions (which all of SWEBench are) to a full-performance variant.
Gareth321 8 hours ago [-]
I feel like a conspiracy theorist but it feels like every new model release from both Anthropic and OpenAI has 1-2 weeks of fantastic performance then a gradual (and sometimes not so gradual) decline in intelligence. It's like they're quantising the model in the background to optimise for available compute/RAM. Which, if I put my MBA hat on, would make perfect sense. Why not halve the required RAM for "90%" of the performance? They get fantastic benchmarks and glowing reviews on release, then slowly squeeze more performance out of the model. By the time the next model is ready for release, the jump feels quite large again.
There's clearly random variation, but it also shows each model release is just genuinely better. ( With the exception of a heavily degraded week of opus 4.7, which was acknowledged as a problem at the time. )
There's a psychology of getting used to models after being wowed by the new performance. It sets in as your new baseline expectations, and then when it doesn't deliver, it's felt more acutely. When it does deliver, it's just meeting expectations.
Then a new better model comes along and it's a step up again, another wow moment for a week or two until expectations adjust to meet the new baseline.
This is also my experience. I don't know if it's because of the quantization theory, or if it's just me getting used to a certain level of coding performance and gradually less tolerant of the mistakes it makes more over time.
ACCount37 5 hours ago [-]
You should. Feel like a conspiracy theorist when saying things like this.
Users are not reliable or consistent model evaluators. Users adapt to models - the moment they get a model that performs better, their expectations rise, their tasks get harder and their prompts get shorter.
"They made the model worse" is PEBKAC in 9 cases out of 10.
jbird99 11 hours ago [-]
At this point I don't even bother with it. Constantly falls back to Opus anyway, so I may as well save myself some time.
Balinares 5 hours ago [-]
Can you elaborate on what felt revolutionary to you about Fable?
sothatsit 48 minutes ago [-]
I got Fable to run overnight and I woke up to a working prototype of a very complex feature. And then I did it again for another complex feature the next night.
The code still took weeks to clean up, but it worked and was correct. It felt then, and still feels, like a big step change on very hard problems. These are problems I would previously expect to take a month or longer to implement.
I have also noticed Fable can handle much more nuance when reasoning through writing and research, but that is harder to quantify.
slopinthebag 30 minutes ago [-]
What was the very complex feature?
dahdum 3 hours ago [-]
Not OP, but to me it initially felt extremely proactive and energetic, just powering through roadblocks with ingenuity and enthusiasm. After it came back I was constantly getting refusals and downgrades for things Opus had been doing. I’ve written it off for my use cases and getting by just fine with Opus 4.8 and now 5.
fastball 14 hours ago [-]
Medium vs High? Why? From all the charts I've seen the performance jump is pretty large from med -> high (not as noticeable from high -> xhigh).
ValentineC 13 hours ago [-]
Medium or low supposedly prevents Opus 5 from overthinking:
Quality vs cost - medium is the sweet (perhaps better too!) spot.
dhorthy 13 hours ago [-]
my issue with frontier code is that it uses a model judge for quality whereas slop code bench forces a model to grapple with its own garbage code in order to receive a functionality reward
sothatsit 14 hours ago [-]
If I need something smarter I use Fable. Medium works well and is quick. Opus 5 medium feels much better to me than Opus 4.8 medium.
dhorthy 14 hours ago [-]
yeah someone will have to re-run this bench on various effort levels. unfortunately it is not cheap
14 hours ago [-]
mellosouls 7 hours ago [-]
This is nice, but would be really useful measured against human performance, and yes - I understand that is likely a difficult challenge.
Many people though are going to read the headline figures and think it means - say - Opus 5 is only a quarter the strength of a human coder.
4by4by4 11 hours ago [-]
I'd be curious to see the raw test results.
I have a suspicion that most models will miss the `database_migration` Checkpoint 2 test that includes a `default_value` because it could be interpreted as either a JSON-literal or a SQL-expression.
There might be other tests as well that are prone to failure for reasons other than the reasons cited in the paper.
I think a cool experiment would be to adjust the order of the features implemented (e.g. checkpoint 3 then 2 then 5 then 4) where dependencies allow it. Then one could account for some checkpoints being more difficult than others.
dhorthy 11 hours ago [-]
oh i really like the idea of flipping around the order of checkpoints and comparing results. Could be an interesting way to increase/decrease difficulty even
i will look into how easy it would be to zip up some subset of the results without leaking anything...probably doable
Johnny_Bonk 16 hours ago [-]
I haven't joined your chats in a while but glad to see you put this together, I truly feel as though opus 5 is not much of an improvement. The only time i ever felt a wow factor was opus 4, 4.6 and fable pre trump admin lobotimizing
dhorthy 16 hours ago [-]
yeah this was just a start - the fastest cheapest thing we could try for a brand new model.
I'm hoping to do some more work with sol/fable in the mix as well as exploring more languages and curating the problem set to include more of the benchmark
I also kinda felt like opus4.5 was dumber than 4.1 personally, maybe a little biased since 4.5 was 2.5x faster and 2.5x cheaper seems to indicate its a smaller model
Johnny_Bonk 15 hours ago [-]
yeah i agree
dan_gee 15 hours ago [-]
[flagged]
scrollaway 15 hours ago [-]
How is this useful or insightful?
You ever go to forums full of entomology specialists and tell them you don’t understand their fancy terms?
dan_gee 15 hours ago [-]
My point is that the differences between these models are so minor that obsessively benchmarking them comes across as navel-gazing.
joatmon-snoo 15 hours ago [-]
The evidence that proves a model is actually a step function change is these benchmarks.
If a model isn’t a step function change? Welcome to research.
13 hours ago [-]
Johnny_Bonk 15 hours ago [-]
like all good science, measure everything
anentropic 4 hours ago [-]
So far my 'solution' to this has been periodically run a separate round of whole-codebase code review (preferably Fable) and then rounds of refactoring off the results of that
thatfunkymunki 2 hours ago [-]
yes, this has been my preferred approach as well. Too much risk of very deep local maxima otherwise.
spicyusername 1 hours ago [-]
It'd be interesting to compare this to human teams, given the same challenges.
swiftcoder 5 hours ago [-]
I wonder to what extent we could steer performance on this benchmark, by providing an adversary model that penalises code duplication and overall lines of code?
killingtime74 15 hours ago [-]
Did you not benchmark latest GPT 5.6 or GLM 5.1/Kimi K3 because of cost? I can run them if you share how you ran them
dhorthy 15 hours ago [-]
no i'm spinning those up at some point this week. here's the first few prompts I used (claude opus 5 as the research orchestrator), (these were interspersed with lots of tools and assistant messages but it should get you kicked off.
> fetch this article for slopcodebench and help me run an eval on a subset of problems with opus 5 https://arxiv.org/html/2603.24755v1
> Get all the context, fetch any mentioned repos, and then propose a plan to me.
> i have an anthropic API key in ....
> Let's do the three challenges with Opus 4.8 and Opus 5 and Fable please. I like your minimal set. Let's try it. What do you need from me?
> Actually I changed my mind. I want to do two of the easy ones you picked and then I want you to pick the one with more checkpoints, maybe one of the harder ones, not the very hardest one but one with a higher number of checkpoints.
> Actually let's do one easy, one medium, and one hard problem please. If we have a hard problem I'd like to see that.
> lets rock - i think lets just do opus 4.8 and sonnet 5 and opus 5 since we our ZDR will block fable
killingtime74 13 hours ago [-]
Thank you!
javed6542 52 minutes ago [-]
Good benchmark. Although, why did they have to give it such a unprofessional name?
ajwin 14 hours ago [-]
To what degree is this a harness/system prompt problem? Models maybe should implement new stuff with as little impact on the existing stuff as possible by default? A simple system prompt for it to always check the code after task completion for proper simplifications, abstractions and cleanups before returning to the user? Instructions to retain "story like" readability of the code.
igregoryca 14 hours ago [-]
At least for Claude Code, putting "run /simplify at the end" in an "implement the plan" skill helps a little. It still often leaves new code in bizarre places, and/or with bad/alien-sounding names and comments.
dhorthy 11 hours ago [-]
Yeah I would hold that models don’t know how to simplify because most rl/benchmarks doesn’t penalize complexity
dhorthy 14 hours ago [-]
I agree this is an option, and the next thing on my radar is to try with a more realistic "factory-shaped" harness where you have feedback from linters and other models after each coding episode that refines the architecture.
For readability specifically, I've found it hard to get the models to do this with prompting. If you've talked to opus/fable for a long time on prose writing you probably felt this too
lostdog 12 hours ago [-]
Clearly the first step of slopbench2 should be to have the agent first write its own harness!
renezander030 6 hours ago [-]
[flagged]
13 hours ago [-]
piazz 10 hours ago [-]
Great writeup. The excessive function thing has always driven me crazy; I guard against this explicitly in Claude.md.
I have found that models are generally poor at managing refactors / complexity while also implementing new features. But I’ve had some success with a semi-lights-off approach where you decompose it and prompt the model adversarially in a second pass to look for new rough edges and areas of complexity or refactors that might simplify the codebase.
So I’d be very curious to see this benchmark but with something like a periodic “refactor turn” interleaved in.
Also eager to see Fable benchmarked; anecdotally that was the only model whose code I felt I could actually trust to not review closely.
insumanth 3 hours ago [-]
Opus models were straightforward to review.
But Opus 5 is very different.
vintermann 5 hours ago [-]
This benchmark makes me worry a bit that people will just ask their model to reimplement everything from scratch once their requirements become more clear.
dhorthy 2 hours ago [-]
why would that worry you?
adamtaylor_13 2 hours ago [-]
I may be crucified for asking this but: is there any proof that slop matters beyond our sensibilities as developers?
If the code is ugly, but defects are low—does it matter?
If the code is hard to read, but clients are happy—should I care?
Genuinely asking. Weird times we live in.
bodeadly 2 hours ago [-]
Like anything in coding, it depends. LLMs are good at following conventions. If what you're doing has been done a 1000 times before and there is a good bias, it will produce good results. This is why LLMs are so good at the leafy parts and even the branches the leaves are connected to. But there's always something unconventional about a codebase. If there weren't, it wouldn't be worth writing.
jdm2212 2 hours ago [-]
It's the eternal question with any kind of tech debt -- is it worth a little more velocity now in return for medium-to-long term slowdown? And there's no general right answer.
dhorthy 2 hours ago [-]
agree, i think the implication is that low quality code is harder to change in the future
paffdragon 1 hours ago [-]
> is there any proof that slop matters beyond our sensibilities as developers?
I've seen a few papers recently on the topic in terms of LLM, here is two I found by quick googling
"Our findings suggest that traditional maintainability principles remain highly relevant in the era of AI-driven development, shaping the computational cost and navigational efficiency of coding agents."
"Our findings confirm that human-friendly code is also more compatible with AI tooling."
"Investing in maintainability not only helps humans; it also prepares for large-scale AI adoption."
> If the code is ugly, but defects are low—does it matter?
> If the code is hard to read, but clients are happy—should I care?
For you who wrote it and are the sole developer, maybe not. But if you want to have other contributors or hand it off, then it can be a problem. New devs joining the project may not want to work with it and will push for a rewrite that will cost money or need to spend extra time on working with code that is hard to work with for them which will cost money. And from the papers above it seems this also affects LLMs as well as they seem to work more efficiently with "cleaner" code.
Majromax 2 hours ago [-]
> I may be crucified for asking this but: is there any proof that slop matters beyond our sensibilities as developers?
That's precisely what this benchmark tries to quantify. Since the benchmark incrementally expands the scope of each problem, 'sloppy' code is code that is hard to later modify.
I know this firsthand: the dumbest coder I've ever worked with was 'myself six months ago'. That jackass never keeps the documentation up to date and hard-codes things that ought to be exposed as configuration.
The SlopCodeBench is an important but early-stage probe in this direction.
willsmith72 10 hours ago [-]
> The big headline is that Opus 5 got a 24% on the small subset of the benchmark that I ran - not much higher than Opus 4.6's 17% strict pass rate in the original paper.
a 41% improvement is not much higher? come on that's just doomer
well_ackshually 9 hours ago [-]
If you tell me that 76% of your code is dogshit but that's it's a massive improvement over your previous 83%, I'm firing you.
1123581321 7 hours ago [-]
I’d enjoy working with such a person, a bleak Eeyore-Gilfoyle.
akurilin 13 hours ago [-]
Really hoping that all of the attention you're bringing to the longitudinal sloppification of codebases makes it back to the labs and creates some pressure to improve that trait of the models. This new benchmark seems promising.
At the same time, I imagine it will be hard for them to prioritize this over improving flashy one-shots of impressive zero-to-one feats that demo so well and attract more customers.
Can we have simonw make "pelican on a bicycle after 1000 requests for iteration" popular?
zparky 3 hours ago [-]
I remember reading a blog post a year or two ago about prompting like a printer driver and asking for increasingly ridiculous changes like time travel or wormhole capability. I don't know if I hallucinated that post but I haven't been able to find it again and I REALLY wish I could
dhorthy 13 hours ago [-]
i laughed at the pelican bit its good
yes the labs will always prioritize the vibeslop dopamine casino as far as I can tell - making the models useful and addictive for unsophisticated users, sometimes at the expense or at the very least at the ignorance of the needs of power users
subarctic 11 hours ago [-]
I still don't have access to Opus 5. I'm on the latest version from Homebrew, I guess the update hasn't made it there yet
sunaookami 9 hours ago [-]
Have you tried running /model claude-opus-5?
sethherr 11 hours ago [-]
Why are you on the version from homebrew?
dhorthy 10 hours ago [-]
somebody get this man a curl-pipe-bash stat
cesarvarela 12 hours ago [-]
Please add Fable; a good benchmark should show that Fable is less prone to just autocomplete and instead pushes back or is at least more tasteful.
dhorthy 12 hours ago [-]
Yeah the main reason I skipped fable was because we have a ZDR with anthropic and I didn’t feel like spinning up another account to circumvent that. Next run will have fable and sol
dcl 16 hours ago [-]
finally the benchmark for me
dhorthy 16 hours ago [-]
i hope that is because you hate slop and not because you write it
weiliddat 8 hours ago [-]
For most of these benchmarks, I feel like a p50 and p95 (using SWE salary as a proxy?) human benchmark as reference would be interesting.
Can we use those metrics in a review pass for every PR? So review agent will have to pinpoint new complexity and propose refactoring or justify increase?
drcongo 6 hours ago [-]
> [...] with Opus 5 writing five times the number of functions/callables than Opus 4.8 over the course of the same set of challenges.
Is this bad? I have McCabe complexity switched on in Ruff and find it a handy watermark for when something should be broken up into smaller, individually testable callables. Five times as many callables could make for much more readable and testable code.
knighthacker 15 hours ago [-]
This is where Opus 5 shines
cute_boi 15 hours ago [-]
Opus 5 is an overconfident stupid model. It tries to generate too much slop, tries to act like everything will fall. I have reversed back to fable and codex sol.
dhorthy 15 hours ago [-]
yes sol is still my daily driver for most coding tasks
I did find opus 5 quite handy for general knowledge work and visual design, without the cost of fable (e.g. the graphics in this post are made by opus 5)
but its not noticeably better than opus 4.8 in those regards, and I would not miss it if forced to go back to 4.8
roncesvalles 14 hours ago [-]
I almost prefer Opus 4.8. Opus 5.0 has the overly scholastic tone of Fable but without the intelligence.
patwolf 13 hours ago [-]
We've been running automated code reviews on claude with a bunch of skills/subagents with different specialties. Any review feedback is then fed back into claude to fix. Since switching to Opus 5 I've noticed the reviews are overly pedantic, and that leads to feedback loops where each fix generates more feedback, which requires more fixes, i.e. slop. I had, for example, a simple SQL migration script with a single CREATE TABLE. After a few rounds of review, it ballooned into a complicated 200 line script.
I'm not ready to blame Opus 5 for being stupid. Perhaps we have a prompt buried somewhere that's essentially asking it to be pedantic, and it's just obeying the prompt.
latentsea 11 hours ago [-]
> I'm not ready to blame Opus 5 for being stupid. Perhaps we have a prompt buried somewhere that's essentially asking it to be pedantic, and it's just obeying the prompt.
This is a known failure mode. Sadly, working on a software engineering team doing agentic engineering now means we need to build and maintain suites of evals that measure these things, so that we can measure the effects of changes to harness components including how they perform under model upgrades.
But... for a traditional software engineering team that has no experience in this... How do we even do it?
tamimio 13 hours ago [-]
So many benchmarks more the models themselves.. just make one unified standard to benchmark all or stop calling it “benchmarking” as this word lost its meaning.
sqemo 11 hours ago [-]
[flagged]
joka88xj 10 hours ago [-]
[flagged]
mengram-ai 3 hours ago [-]
[flagged]
delbertty 13 hours ago [-]
Mostly a harness problem in my experience. Slop accumulates when the agent can touch anything, so constraining it to one seam and having it add alongside rather than edit in place does more than model choice.
I’ve used SCB as part of my assessment of agent skills (superpowers, GSD etc) https://orcabot.com/labs/do-skills-improve-coding-agent-accu...
There is a small but growing community on discord for discussing SCB so if interested please join https://discord.gg/BrC4BA9sVj
What is also nice about this benchmark is that it can be used to iterate on prompts/skills for reducing code complexity.
Labs do not train on benchmark data (allegedly). They can train on similar problems, but benchmarks have specific strings in them that labs are supposed to be aggressive in filtering out of their training corpora.
The original paper cited by this post does try to see if improved prompting will make a big difference in the end using a `plan_first` prompt variant, but find no influence on pass rate at the end of the benchmark. The `plan_first` seems to assume coding agents will just refactor once they finish features, but I don't think they tend to refactor significantly unless they are told to fix bugs rather than build features. The benchmark leaves tests hidden with no fail-to-pass feedback, so that may be why degradation is monotonic.
It's especially relevant now that models are good enough to solve ~most point-in-time problems.
Some relevant but disconnected thoughts:
- deterministic scores are so nice
- what "maintainable" is is probably some high dimensional space described by these signals; it'd probably require some human labeling to figure out where this space is
- another signal I've been thinking about and I'm seeing increasingly get brought up is the state space of a system; I'm seeing formal methods pop up a lot recently
State space of a system AND the way to make it accessible / visible to a model. Many times a model can work magic if it can "see" the state of a system in a way that suits it. That's why sometimes having a cli added to the environment seems like such a big unlock. Because that cli usually takes a complex state and allows visibility into it, and possible manipulation in a structured way.
I think the problem with just encoding a whole plan in a single markdown file is that it gets polluted really quickly (agents can stop adhering to instructions to keep it clean and conform to a specific structure), which makes it harder for agents to see which parts of the plan they should focus on. As such, I've reached a similar conclusion about giving the agent access to CLI tools to help them deal with this. To try to mitigate this, I've been getting Claude to develop for me a CLI tool that:
1. Scaffolds reusable, structured plan templates and a reusable workflow that structures how to tackle the plan, step by step.
2. Validates that the plan files still conform to the correct, parseable structure.
3. Parses and evaluates the plan files, to determine what the current state is and what the next valid transitions and states are according to the workflow, like a state machine, and outputs instructions and reminders for agents as to what they should do at each step of the workflow.
So far, I've been dogfooding the tool and it seems promising: I can leave Claude running for longer and it doesn't drift as much. However I haven't ran any benchmarks yet and I'm still not entirely happy with the state of the codebase ( https://github.com/nothingnesses/agent-scaffold ), so take this with a grain of salt.
That said, I'm also bullish on using agents with formal methods and proofs. Type checkers and compile-time checking in general are great because they surface errors early and with great specificity. So if you can encode your specifications with, e.g. dependent types, you can use the type-checker as a way to steer the agent when it goes wrong and gets off-track.
this is a nicely succinct way to put this - a multi-dimensional space where no single metric is really useful
state space of the system is interesting too. I would guess that for any production software with dependencies like databases/third parties that might be too hard to measure, but if you can silo off parts of your system into bounded state machines, it may be a value metric on some module behind a clean interface.
I think the kubernetes control loop model is a great instance of this, a handful of scoped components that own a control loop across a well-defined state machine, that can operate / recover in the face of most network partitions or downtime - the promise of CRDTs but rather more a pragmatic approach to it
I’ve now replaced my use of Opus 4.8 xhigh with Opus 5 medium, and I’m using less tokens and it’s quicker. I can understand people being annoyed by its writing style but for getting work done that really doesn’t bother me. I’ve been really enjoying using it.
It seems like there’s a week by week and sometimes day by day change in performance when on a subscription plan using their harnesses.
There's clearly random variation, but it also shows each model release is just genuinely better. ( With the exception of a heavily degraded week of opus 4.7, which was acknowledged as a problem at the time. )
There's a psychology of getting used to models after being wowed by the new performance. It sets in as your new baseline expectations, and then when it doesn't deliver, it's felt more acutely. When it does deliver, it's just meeting expectations.
Then a new better model comes along and it's a step up again, another wow moment for a week or two until expectations adjust to meet the new baseline.
Users are not reliable or consistent model evaluators. Users adapt to models - the moment they get a model that performs better, their expectations rise, their tasks get harder and their prompts get shorter.
"They made the model worse" is PEBKAC in 9 cases out of 10.
The code still took weeks to clean up, but it worked and was correct. It felt then, and still feels, like a big step change on very hard problems. These are problems I would previously expect to take a month or longer to implement.
I have also noticed Fable can handle much more nuance when reasoning through writing and research, but that is harder to quantify.
https://xcancel.com/danshipper/status/2080700057892815114
Quality vs cost - medium is the sweet (perhaps better too!) spot.
Many people though are going to read the headline figures and think it means - say - Opus 5 is only a quarter the strength of a human coder.
I have a suspicion that most models will miss the `database_migration` Checkpoint 2 test that includes a `default_value` because it could be interpreted as either a JSON-literal or a SQL-expression.
There might be other tests as well that are prone to failure for reasons other than the reasons cited in the paper.
I think a cool experiment would be to adjust the order of the features implemented (e.g. checkpoint 3 then 2 then 5 then 4) where dependencies allow it. Then one could account for some checkpoints being more difficult than others.
i will look into how easy it would be to zip up some subset of the results without leaking anything...probably doable
I'm hoping to do some more work with sol/fable in the mix as well as exploring more languages and curating the problem set to include more of the benchmark
I also kinda felt like opus4.5 was dumber than 4.1 personally, maybe a little biased since 4.5 was 2.5x faster and 2.5x cheaper seems to indicate its a smaller model
You ever go to forums full of entomology specialists and tell them you don’t understand their fancy terms?
If a model isn’t a step function change? Welcome to research.
> fetch this article for slopcodebench and help me run an eval on a subset of problems with opus 5 https://arxiv.org/html/2603.24755v1 > Get all the context, fetch any mentioned repos, and then propose a plan to me.
> i have an anthropic API key in .... > Let's do the three challenges with Opus 4.8 and Opus 5 and Fable please. I like your minimal set. Let's try it. What do you need from me?
> Actually I changed my mind. I want to do two of the easy ones you picked and then I want you to pick the one with more checkpoints, maybe one of the harder ones, not the very hardest one but one with a higher number of checkpoints.
> Actually let's do one easy, one medium, and one hard problem please. If we have a hard problem I'd like to see that.
> lets rock - i think lets just do opus 4.8 and sonnet 5 and opus 5 since we our ZDR will block fable
For readability specifically, I've found it hard to get the models to do this with prompting. If you've talked to opus/fable for a long time on prose writing you probably felt this too
I have found that models are generally poor at managing refactors / complexity while also implementing new features. But I’ve had some success with a semi-lights-off approach where you decompose it and prompt the model adversarially in a second pass to look for new rough edges and areas of complexity or refactors that might simplify the codebase.
So I’d be very curious to see this benchmark but with something like a periodic “refactor turn” interleaved in.
Also eager to see Fable benchmarked; anecdotally that was the only model whose code I felt I could actually trust to not review closely.
If the code is ugly, but defects are low—does it matter?
If the code is hard to read, but clients are happy—should I care?
Genuinely asking. Weird times we live in.
I've seen a few papers recently on the topic in terms of LLM, here is two I found by quick googling
* "Does Code Cleanliness Affect Coding Agents? A Controlled Minimal-Pair Study" https://arxiv.org/abs/2605.20049
"Our findings suggest that traditional maintainability principles remain highly relevant in the era of AI-driven development, shaping the computational cost and navigational efficiency of coding agents."
* "Code for Machines, Not Just Humans: Quantifying AI-Friendliness with Code Health Metrics" https://arxiv.org/abs/2601.02200
"Our findings confirm that human-friendly code is also more compatible with AI tooling."
"Investing in maintainability not only helps humans; it also prepares for large-scale AI adoption."
> If the code is ugly, but defects are low—does it matter? > If the code is hard to read, but clients are happy—should I care?
For you who wrote it and are the sole developer, maybe not. But if you want to have other contributors or hand it off, then it can be a problem. New devs joining the project may not want to work with it and will push for a rewrite that will cost money or need to spend extra time on working with code that is hard to work with for them which will cost money. And from the papers above it seems this also affects LLMs as well as they seem to work more efficiently with "cleaner" code.
That's precisely what this benchmark tries to quantify. Since the benchmark incrementally expands the scope of each problem, 'sloppy' code is code that is hard to later modify.
I know this firsthand: the dumbest coder I've ever worked with was 'myself six months ago'. That jackass never keeps the documentation up to date and hard-codes things that ought to be exposed as configuration.
The SlopCodeBench is an important but early-stage probe in this direction.
a 41% improvement is not much higher? come on that's just doomer
At the same time, I imagine it will be hard for them to prioritize this over improving flashy one-shots of impressive zero-to-one feats that demo so well and attract more customers.
Can we have simonw make "pelican on a bicycle after 1000 requests for iteration" popular?
yes the labs will always prioritize the vibeslop dopamine casino as far as I can tell - making the models useful and addictive for unsophisticated users, sometimes at the expense or at the very least at the ignorance of the needs of power users
Edit: FWIW the paper the post quoted has repositories as slop baseline https://arxiv.org/html/2603.24755v1#S4.SS2
Is this bad? I have McCabe complexity switched on in Ruff and find it a handy watermark for when something should be broken up into smaller, individually testable callables. Five times as many callables could make for much more readable and testable code.
I did find opus 5 quite handy for general knowledge work and visual design, without the cost of fable (e.g. the graphics in this post are made by opus 5)
but its not noticeably better than opus 4.8 in those regards, and I would not miss it if forced to go back to 4.8
I'm not ready to blame Opus 5 for being stupid. Perhaps we have a prompt buried somewhere that's essentially asking it to be pedantic, and it's just obeying the prompt.
This is a known failure mode. Sadly, working on a software engineering team doing agentic engineering now means we need to build and maintain suites of evals that measure these things, so that we can measure the effects of changes to harness components including how they perform under model upgrades.
But... for a traditional software engineering team that has no experience in this... How do we even do it?