I suggest reading the Mythos report's discussion on SWE-bench and contamination. I think it's fairly convincing that you can account for contamination and still trust SWE-bench numbers on models that aren't over-optimized for it.
replyYou can trust that a model that scores 40% vs a model that scores 90% is indeed worse.
replyYou can’t trust it that a model that scores 93% is better at software engineering than a model that scores 90%, because at that point it’s impossible to distinguish between recall and reasoning.
It’s honestly far better to just ignore SWEBench Verified in 2026. Multiple labs have noted issues with contamination, and achieving high scores require memorisation of what passes the prescriptive verifier; not what is a correct solution.
reply40% vs 90%? Sure.
70% vs 90%? _Absolutely meaningless_ as you are not measuring coding intelligence but “how well can the model cheat flaws in SWEBench Verified”, the former can certainly be better at coding even assuming no deliberate benchmaxxing / foul play.
> models that aren't over-optimized for it.
replyBut how do you know the model was over-optimized for it or just really good?
reply
I don't understand that methodology in the first place. Does Anthropic even have some kind of somewhat objective definition to measure and judge "memorization"? Is there any evidence that other LLMs are viable tool to determine that?
replyThis article says anthropic models can write out the entire benchmark solution set word for word from memory
replythere's more details under the Too narrow and too wide tests heading.
replyIt would be interesting to see a deeper investigation, into how the models are dealing with this and whether the successful ones seemed to be trained on the benchmark.