Separating signal from noise in coding evaluations

https://openai.com/index/separating-signal-from-noise-coding-evaluations/

by sk4rekr0w • 15 days ago

239 97 15 days ago

openai.com

2001zhaozhao• 15 days ago

Translation: other labs have learned to benchmaxx SWE-Bench Pro better than they do

xacky• 15 days ago

Achieving AGI will be more than just passing all benchmarks, it has to account for the unknown problems too.

metalliqaz• 15 days ago

Unless they have something in the labs that massively departs from their current products, AGI isn't on the table and is purely hype for marketing purposes.

cyanydeez3• 15 days ago

they should be consulting Donald Rumsfeld and make sure they implement the Unknown-Unknowns benchmark, because thats how they get you

minimaxir• 15 days ago

This ties into the bias-variance tradeoff (https://en.wikipedia.org/wiki/Bias%E2%80%93variance_tradeoff) common with building non-LLM models. The solutions can only be a) figure out how to get LLMs smaller with similar performance so they don't memorize things/game the benchmarks and b) build benchmarks that are indeed comprehensive for all real-world data, which is infeasible.

sigbottle• 15 days ago

I mean, people always say there are tradeoffs, until you reach the next frontier, in which there are tradeoffs at said frontier, and the next, and the next, etc.

In one sense, yes, tradeoffs are inescapable as the scope expands to the maximal possible scope. In another sense... it depends on the level of abstraction we're talking about.

naikrovek4• 15 days ago

AGI is a long way off. Unless you’re talking about some unknown-to-me LLM marketing BS which is called “AGI” or something, I guess. Artificial general purpose intelligence is so different to LLMs or image AI that they are completely incomparable, except to say that they are all artificial. AGI will do a lot more than token prediction.

ACCount374• 15 days ago

What's your evidence of that? That AGI requires a truly novel architecture, and not just another iterative "LLM but with an extra trinket and wheels that spin ten times faster".

theLiminator2• 15 days ago

Please define AGI first.

bellowsgulch3• 15 days ago

Seems like depending on your field these days, the hot thing to do is build your own private benchmarks.

In my own testing, no frontier model knows how to replicate an original 1990s Super Soaker prototype design, which for the most part, should be almost completely possible with Home Depot parts.

They just don't understand PVC parts, triggers, etc.

softwaredoug• 15 days ago

Or defensively expect models to be stupid.

Seems the smart thing to do is not assume an agent will do the right thing. But to create the scaffold / harness that enforces constraints to steer them towards a good result.

Then you can swap out the really smart model for maybe something cheaper.

thierrydamiba• 15 days ago

Or you’re getting steered into la la land because of your prompt

bellowsgulch3• 15 days ago

Certainly, but deconstructing the problem, none of the models seem to appreciate the staggering difference between a ball valve and a button release.

Of course, there's also no super soaker engineer jobs to take, so I'm sure training sophisticated models to do well in that area is not a high priority for any firms.

ACCount374• 15 days ago

It's a "commonsense spatial reasoning/problem solving" kind of problem. LLMs fail at spatial reasoning forever.

What humans "easily" solve in seconds with raw spatial reasoning LLMs often find easier to solve by invoking A* or a constraint solver.

Might be that text data is particularly bad at teaching that to LLMs. Or that being good at spatial reasoning requires true recurrence, and autoregressive chain of thought is a poor substitute. Or it might be that human brain was optimized by evolution for solving spatial problems in open ended 3D environments for hundreds of millions of years, optimized for language for mere hundreds of thousands of years, and only optimized for writing computer code for a few decades at most.

The current frontier is halfway competent at benign closed 2D work, but still completely fumbles anything remotely close to open ended real world 3D work. It's getting better, but very slowly.

therobots9276• 15 days ago

A literal bird brain would outperform an LLM on most spatial reasoning tasks.

Extrapolating the core theory of LLMs - that we can reverse engineer reasoning through language - does that imply that if we train a bird song LLM to predict next “token” (pitch) of a birdsong, that the LLM could excel in a bird flight simulator?

I think it’s pretty clear that this is a dead end.

ReptileMan2• 15 days ago

Lately my benchmark is build123d - trying to force them to build me functional parts only by the description. All of the models don't perform well.

mgiampapa• 15 days ago

IDK, sounds like it has brute forced my password already.

midtake• 15 days ago

This guy builds

cyanydeez3• 14 days ago

was watching some youtube about the text/vision models and the trouble with language as a descriptor; their novel idea was to use coordinate systems on visual imput so the model would map out an image, then it tags what's looking at with coordinates or boxes, and then think with those box coordinates, providing a level of disambiguation.

If you tried to explain the same stuff using purely text, you'd also need to come up with some kind of "language", which you know is programming.

So 1: you'd need to fine tune a model to actuall succeed; they're not reaching AGI any time soon with the current batch. 2: you need to develop a lingual DSL for it, as they'll never do much of anything without some kind of glue and disambiguation.

ReptileMan2• 14 days ago

This is why cad cam are such a good test at the moment.

shay_ker• 15 days ago

Didn't we all know from the start that all of SWE-Bench was flawed? Even the authors concede the limitations and have long since moved on.

paxys• 15 days ago

SWE-Bench Pro was created to replace SWE-Bench and fix these problems.

warkdarrior• 15 days ago

SWE-bench Verified was created to fix the problems of SWE-bench.

Then SWE-Bench Pro was created because SWE-bench Verified had flaws.

Now SWE-Bench Pro is shown to have flaws.

dandaka2• 15 days ago

What is considered SOTA for SWE benchmarks now?

EuanReid• 15 days ago

I've generally found DeepSWE[0] to be pretty true to reality.

[0] https://deepswe.datacurve.ai/

cheesecakegood• 14 days ago

Oh very interesting, I didn’t realize I should probably be using Fable Medium more than High, due to how that curve and the cost looks!

enraged_camel• 15 days ago

FrontierBench

dandaka2• 15 days ago

do they have a website? I have found only paper PDF and it seems more general than SWE

carabiner3• 15 days ago

strawberry

retr0rocket• 15 days ago

Why is this a problem? Its like asking a person how many elder futhark runes are in the word strawberry.

Unless you want to tack on bpe enconding table to every llm context its pointless

swyx• 15 days ago

https://cognition.ai/blog/frontier-code (disclaimer - was on the team - but also we covered swebench pro/deepswe issues in here as well.)

sk4rekr0w• 14 days ago

1.1 seems a lot better than the original release, which was a bit hyperbolic. excited to see the team keep iterating.

Topfi2• 15 days ago

Either DeepSWE [0] or FrontierCode [1], depending on personal goals and requirements. The later is more interesting for me personally, due to the design of the benchmark heavily grading "mergability", i.e. how the provided output is to review and whether a serious developer can easily parse it and'd be willing to merge the result. In my mind and with my private evals, for quite some time I've held firm that a model can have a higher ceiling but that has limited value if I do not feel truly confident in signing off on the code.

[0] https://deepswe.datacurve.ai/

[1] https://cognition.com/blog/frontier-code-1.1

codedokode2• 14 days ago

Sadly no tasks for C, especially for working with optimized low-level data structures. And with testing performance of a solution. Anyone can write Python, try writing optimized low-level code.

Also I wonder if models playing dumb to prevent learning on outputs affected the score.

Also interesting that Claude edits files by writing and running Python scripts, is that efficient?

porphyra• 15 days ago

Interesting timing to release this just when SWE-1.7 and Grok 4.5 came out being much cheaper than GPT-5.5.

johngoode• 15 days ago

This doesn’t seem like opportune timing to announce days before a new model drop

janalsncm• 15 days ago

Based on the numbers here it seems there’s less than 800 tasks in the entire benchmark. That is enough for a handful of engineers to comb through in a week (which is what OpenAI eventually did here).

On the one hand, kudos to them for actually doing that work.

On the other hand, garbage in, garbage out. It’s a bit embarrassing for the original authors to have not actually checked, and it’s embarrassing for everyone downstream to have not checked either.

Also if you check the article, although an LLM did find issues, it tended to underestimate issues that professional software engineers found.

jheitmann• 15 days ago

It reads to me like "We did all the work you'd do to figure out how to fix the benchmark, then we decided to throw out the benchmark". Is there some reason the underlying data is so golden that it can't be patched? At the end they argue for a slightly more curated approach to benchmark generation, but my gut is that using messy ill-specified tests taken from real world data and patching them into fairness would be a pretty solid path to take.

Centigonal• 15 days ago

If they fixed it, then it wouldn't be SWE-Bench Pro anymore, right? It'd be "SWE-Bench-Pro-Fixed-OpenAI." I think it's better optics for the independence of the benchmark if the OpenAI team lets some third party do the fixing and release the improved benchmark.

...Although OpenAI did exactly that when they released SWE-Bench Verified, so maybe I'm talking out of my butt here.

kimjune01• 2 days ago

eval shops who depend on these things for a living will have to version their datasets, to prevent contamination

tedsanders• 15 days ago

Pointing out problems (e.g., hidden tests that assume narrow implementation details) is much easier than fixing them (e.g., creating tests that work for any possible choice of implementation).

ebcode• 14 days ago

And it reads to me like they have some other reason to move on from SWE Bench Pro, but they don't want to say what it is. They say right up top, "~30% of the tasks are broken." But that leaves ~70% un-broken, which seems pretty good to me. It would be nice if they would also say: "Here's the list of instances that are broken: <CSV>". Or, "Here's the subset of SWE Bench Pro we will use going forward." They're letting the perfect be the enemy of the good.

esperent2• 14 days ago

I think you can be sure they would have done that if it showed their model on or very close to the top.

mlhpdx• 15 days ago

Fundamentally aren’t they concluding that tasks assigned to software developers (human or otherwise) are often incomplete, self contradictory or worse? This is the world in which their tool must play. I’m unsympathetic.

kakugawa• 15 days ago

The more subtle point is that there's a gap between the task and its verification. e.g. if you have an open-ended / under-specified prompt, the verification needs to be able to handle all potential solutions.

So you can have a very narrow task prompt that's easy to verify (but likely too simple of a challenge). Or a more realistic task prompt that's much harder to verify. And likely harder to both build the robust verifier and run it cheaply.

donw• 15 days ago

A substantial portion of software engineering -- and the fundamental jobs of a proper Product Owner and UX Designer -- is to turn "vague ideas about what we need to do" into "this widget, on this page, it should work like this"

It's not a pipeline, it's an ongoing conversation within any functional team, but this requires buy-in from management, who is often selected for "line must go up this quarter no matter the cost" over "hey, wouldn't it be cool if this company was still a going concern in twenty years?"

gilfaethwy• 15 days ago

Agreed - "underspecified prompts" being listed as a failure of the tooling is not a strong case. Even interns can understand ambiguous asks with a bit of help, and understand when they need to stop and ask instead of just carrying on. They are often working fairly independently on ambiguous tasks before the end of an internship, too.

So is the argument that frontier models are not just junior engineers, but first-month interns with no capability of progressing beyond that level?

andy12_• 14 days ago

Even interns can understand ambiguous asks with a bit of help

This is not a case of an ambiguous task. This is literally trying to judge a model based on information it cannot possibly know, like trying to judge someone based on whether they know what I have hidden in my backpack. In the real world an intern could look at unit tests or ask for feedback, but that is not the case in a benchmark.

rao-v• 14 days ago

In a real job, you would be allowed to see the test case that failed and tweak your code (or more likely the poorly written test).

If you let a modern LLM do even the first, they’d crush this specific benchmark.

What is interesting is understanding how LLMs are able to beat 70+% on this benchmark or getting some of the poorly framed questions right? Are they implicitly learning the test writers style? Are the solutions leaking into their training set?

Perhaps reassuring is that even Fable stalls out at ~72% (on the hidden set which OpenAI did not run this analysis on), so perhaps training on the bench is not happening in anything but the most indirect ways.

I care a lot because small open models can never learn idiosyncrasies like this, so I really want good ways to judge models fairly.

EDIT: Humm OpenAI is muddying the water a bit. Only 20%ish of problems are broken in ways that are unfair to the agent, 4-10% are broken in favorable ways, so the benchmark ceiling is probably closer to 80-85%

mkozlows• 14 days ago

This was also my thought, and I think holds true of the ones with invisible requirements that aren't stated up front and are only captured in tests. Oh, you need to rework your solution to handle requirements nobody mentioned before? Well, me too.

Yeah, it's testing a different thing than what the benchmark claims to test, but it's also accidentally testing something more real-world applicable than a clean benchmark would be, so hey.

(EDIT: That is, if the agent is allowed to see the failed tests and iterate. If not, then yeah, that's just a problem. And either way, the ones with tests that just encode a particular solution's implementation details, thereby demanding that your solution have some rando internal details, are junkier. That's not a situation you'd run into in reality.)

CSMastermind• 15 days ago

DeepSWE is the one I generally trust: https://deepswe.datacurve.ai/

therobots9276• 15 days ago

Aren’t we past the point of needing benchmarks? If we’re as close to AGI as Sam says then the proof should be in the pudding. OpenAI should build a competing CRM / Figma / Photoshop with a couple dozen engineers and a Dyson sphere’s worth of compute and just prove the capabilities.

This all feels like a 2024 re-run. Oh, ChatGPT is going to cure cancer? Then find ONE rare cancer and CURE IT. OpenAI has access to the best models and compute - so cure fucking cancer! What the fuck are you waiting for?

whackernews• 14 days ago

Yeh I totally agree. I’d go one step further and say even if one of these models cured cancer it’s still only going to have done it in a computer-y way doing computer stuff. Can that same model experience a beautiful landscape and convey its emotions in an evocative way, could it tell how you’re feeling when you come back from work after a hard day? Could it hop on one leg? What the hell even is AGI and how does it differ from GI? I don’t know what we’re even talking about any more!

therobots9276• 14 days ago

If OpenAI cures ANY form of cancer then I will admit I was wrong.

Until then all we have is a lot of hot air coming out of Sam and Dario’s asses

GodelNumbering• 15 days ago

There are also a lot of fake results out there on Terminal Bench 2 for different reasons (although the great team behind it Ryan/Alex et al, recently cleaned up a lot of dodgy submissions). A lot of labs publish the results by modifying timeouts or hardware config which effectively bypasses what is being tested in certain tasks. Then there is harness level cheating, models reward hacking and more...

In fact, one thing that still bothers me after months is the gpt-5.5 official submission. This task in particular https://www.tbench.ai/leaderboard/terminal-bench/2.0/codex/0...

The task has the following timeouts (https://github.com/harbor-framework/terminal-bench-2/blob/ma...).

[verifier]

timeout_sec = 1200.0

[agent]

timeout_sec = 1200.0

[environment]

build_timeout_sec = 600.0

Which means no agent should take more than 3000 seconds doing it. Two out of five attempts in the link above took well over 3000 seconds (75min and 80 min respectively). Even though they failed, the fact that they ran that long is sus.

Goodhart’s Law at work

bjackman• 14 days ago

Even if nobody is "cheating" your particular definition of cheating, the benchmarks are _somewhere_ in the super-structural gradient descent. Models are benchmark-maximising machines at some level, so I think the benchmarks are inherently a bit useless.

This is not really surprising, benchmarking _people_ doesn't work. You can only get a decent measure of someone's coding abilities by personally interacting with them. Given that models are basically person simulators it would be weird if benchmarks kept being useful as the simulation got more accurate.

I think what I've just said is basically just a more roundabout way of what you said: "Goodhart's law at work". It really is a law.

jjcm2• 15 days ago

I want a new bench - given $100 of api spend, how much can a model accomplish for a suite of benchmark tests?

Give us something that measures a combination of efficiency and intelligence.

I think this would allow for some interesting tactics for smaller models - eg they could do things like computer use to test their results and grind on problems for longer to verify the outputs, whereas larger models may not have budget to self-test.

therobots9276• 15 days ago

This is the fundamental question and don’t you find it interesting that there isn’t a nice clean dashboard on the openAI website where we can go and see this metric progress over the release history?

Toby Ord did what he could with public data and it… doesn’t look great.

https://www.tobyord.com/writing/hourly-costs-for-ai-agents

SyneRyder2• 15 days ago

Seems like you're asking for the Artificial Analysis "Intelligence vs Cost" benchmark, perhaps?

https://artificialanalysis.ai/?cost=intelligence-vs-cost-per...

jjcm2• 15 days ago

Not quite. These cost-per-task benchmarks report the cost of the task after the model gives its initial answer. The total cost is irrelevant, and isn't factored into the model's decisions - a run of the full benchmark for something like Fable might cost $10k.

What I'm looking for is the inverse. I want to give the model a budget of $100, and see how much it can accomplish with that $100. For smaller models, this means they can do more than just choose thinking amount, they can do something like a /loop to keep iterating on a problem until they get it right.

Can something like Deepseek V4 Flash get more answers correct than Fable, when given equal budgets?

Think of it as answering this question: How much intelligence can you get out of a model given a budget of $100? A cost-per-task dash correlates, but it doesn't give you an answer to that question.

schaefer• 15 days ago

It's still spooky to see exponential scales on the money axis.

I do not have exponential funds in my allowance...

cadamsdotcom• 14 days ago

At whose prices?

Ancalagon• 15 days ago

Studying for leetcode exams in the age of AI agent coding evaluations is a wild feeling.

jumploops• 15 days ago

All of the benchmarks are pretty terrible when you look under the hood.

For context, I've been iterating on a "supervisor" to replace a lot of the rigamarole spent when working with Codex/Claude Code, and recently ran this agent against Terminal Bench 2.1

At first I was excited, because my spec-driven supervisor outperformed vanilla codex on a bunch of tasks, however as I looked deeper, I found a ton of issues with the tasks themselves.

The main takeaway is that the instructions are often ambiguous while the test cases are overly specific.

A few examples:

- For `configure-git-webserver` the task includes language like "so that I can run" which blurs the line between what the agent should deliver vs. what should be removed. This causes an overthinking agent to configure the server, and then remove the exact files that the verifier checks, because if the user were to run the same commands, they would conflict.

- For `make-mips-interpreter` the task includes the language "I will check that you booted doom correctly" which causes the agent to retain the generated file `/tmp/frame.bmp` because the supervisor expects the user to check that _it_ booted Doom correctly, not that Doom boots correctly in an isolated way. The verifier then fails to start Doom, because it exits when an existing `/tmp/frame.bmp` exists, not checking to see that it's created from the boot[0].

- For `mcmc-sampling-stan` the supervisor agent often reached the right value, but produced a domain-specific numeric output in scientific notation, rather than a simple decimal form. The verifier fails because it parses the result incorrectly[1].

These are just a few of the inconsistencies I've found, which leads me to believe that Terminal Bench 2.1 is already saturated, and the results from GPT-5.6 and Mythos are basically at the top of the expected threshold (88.8% and 88% respectively).

The biggest issue, as I can tell, is that most benchmarks are "one-shot" and rarely test the model+harness on long iteration tasks, which is the primary way most users use these tools in practice.

[0] https://github.com/harbor-framework/terminal-bench-2-1/issue...

[1] https://github.com/harbor-framework/terminal-bench-2-1/issue...

kasince2k• 15 days ago

we def need a benchmark for all these benchmarks

zvolsky• 14 days ago

The misleading prompt cases are inadvertently testing the model's ability to filter out noise from its instructions. That could be a benchmark on its own. The correct response is to flag the inconsistency and ask for clarification.

jimberlage• 14 days ago

I can understand why this might make for a bad benchmark, but

Overly strict tests enforce specific implementation details not specified in the prompt, invalidating many functionally correct submissions.

Underspecified prompts omit requirements that hidden tests enforce and that are not reasonably inferable.

Low-coverage tests under check the requested feature, so incomplete fixes can pass.

A misleading prompt points models toward the wrong behavior or contradicts what tests require.

If the goal is, "how does my model compare to real SWEs", these are pretty reasonable situations that your model will have to encounter. It's a little like making a nursing exam and then flagging that some of the tests required you to ask the attending doctor for additional information that's not in the chart, or that the patient's family didn't fully explain their aging grandma's medical history.

I can understand why they might want a tighter benchmark, but if you're OpenAI and you promised your model as a replacement for real workers, this isn't the best look. It seems like you would want to test these things.

Separating signal from noise in coding evaluations

Related Stories