--- title: The Programming Languages AI Still Gets Surprisingly Wrong description: AI coding assistants generate syntactically plausible code across nearly every language, but fluency varies sharply by how heavily represented a language is in... canonical: https://daily.dev/posts/the-programming-languages-ai-still-gets-surprisingly-wrong-lbxoqyzy8 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: The Programming Languages AI Still Gets Surprisingly Wrong | daily.dev og:description: AI coding assistants generate syntactically plausible code across nearly every language, but fluency varies sharply by how heavily represented a language is in... og:url: https://daily.dev/posts/the-programming-languages-ai-still-gets-surprisingly-wrong-lbxoqyzy8 og:image: https://api.daily.dev/og/posts/lBXOqYZY8.png og:image:alt: The Programming Languages AI Still Gets Surprisingly Wrong og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # The Programming Languages AI Still Gets Surprisingly Wrong **[Code Like A Girl](https://daily.dev/sources/colkgirl)** · 14 min read · 4 upvotes · 1 comments ## Summary AI coding assistants generate syntactically plausible code across nearly every language, but fluency varies sharply by how heavily represented a language is in training data and how strict its compiler and memory model are. Python benefits from massive representation and forgiving syntax, but studies like EffiBench-X show it also yields more efficient code than Java, C++, or Go, while DeepSeek-R1 posted similar correctness on C++ and Python but notably better efficiency on Python. C and C++ allow memory bugs that compile cleanly and surface only under specific conditions. Rust's strict borrow checker exposes mistakes immediately, but a HackerNoon benchmark found the Python-Rust gap wasn't statistically significant, with Elixir showing the real gap instead; RustEvo² found a knowledge-cutoff effect where models scored 56.1% on pre-cutoff APIs versus 32.5% on newer ones. SWE-bench Pro shows frontier models scoring near 23% versus 70-80% on SWE-bench Verified, highlighting how benchmark design affects perceived capability. The piece recommends specifying language versions, stating constraints upfront, and treating models as collaborators to push back on rather than oracles. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why does Rust seem harder for AI coding assistants than Python even though the correctness gap isn't that large? The perceived difficulty largely comes from the Rust compiler being an unforgiving second reviewer that rejects code immediately, making errors more visible, not from a dramatically higher error rate. A HackerNoon comparative study testing models on 100 problems from October 2025 to February 2026 found the Python-to-Rust performance gap wasn't statistically significant, while Elixir showed a much larger, clearly significant gap. _Anyone weighing Rust against other languages for AI-assisted work can track benchmark writeups like this on daily.dev._ ### How much does an AI model's knowledge cutoff affect its accuracy with newer Rust APIs? Models averaged 56.1% success on Rust APIs available before their training cutoff versus only 32.5% on APIs introduced after, according to the RustEvo\u00b2 benchmark, which evaluated 588 API changes across the Rust standard library and third-party crates. Giving the model retrieval access to current documentation improved performance on newer APIs by an average of 13.5%. _Developers tracking fast-moving crate APIs can follow benchmark research like RustEvo\u00b2 on daily.dev._ ### What's the difference between SWE-bench Verified and SWE-bench Pro scores for AI coding models? SWE-bench Verified scores climbed into the 70-80% range for top models by early 2026, but SWE-bench Pro, a harder variant designed to resist memorization and pattern-matching, shows even frontier models scoring closer to 23%. The gap illustrates that benchmark difficulty and realism matter enormously when interpreting a single headline pass-rate number. _Anyone comparing coding model claims can weigh benchmark nuances like this via daily.dev before trusting a headline score._ ## Community discussion Top comments from developers on daily.dev. **@andradei** · 0 upvotes > Rust makes sense when you put AI agent into a loop to fix it's own errors. Python is the interesting one, I thought Go would score higher for being compiled. ## Similar posts on daily.dev - [AI and machine learning outside of Python](https://daily.dev/posts/ai-and-machine-learning-outside-of-python-jykbfnjcf) · InfoWorld · 0 upvotes · 0 comments - [Why care about programming languages](https://daily.dev/posts/why-care-about-programming-languages-1gwkcjiey) · Lobsters · 2 upvotes · 0 comments - [Will AI force code to evolve or make it extinct?](https://daily.dev/posts/will-ai-force-code-to-evolve-or-make-it-extinct--zwb63chyi) · The New Stack · 3 upvotes · 0 comments - [TypeScript, Python, and the AI feedback loop changing software development](https://daily.dev/posts/typescript-python-and-the-ai-feedback-loop-changing-software-development-uhkhbhbhr) · GitHub Blog · 4 upvotes · 0 comments --- Tags: [#python](https://daily.dev/tags/python), [#rust](https://daily.dev/tags/rust) [View this post on daily.dev](https://daily.dev/posts/the-programming-languages-ai-still-gets-surprisingly-wrong-lbxoqyzy8) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"The Programming Languages AI Still Gets Surprisingly Wrong","url":"https://daily.dev/posts/the-programming-languages-ai-still-gets-surprisingly-wrong-lbxoqyzy8","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/the-programming-languages-ai-still-gets-surprisingly-wrong-lbxoqyzy8"},"datePublished":"2026-08-25T02:11:40.272Z","dateModified":"2026-09-14T07:07:11.105Z","description":"AI coding assistants generate syntactically plausible code across nearly every language, but fluency varies sharply by how heavily represented a language is in...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4f35a01a4fd407dd919724b69a583a37?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4f35a01a4fd407dd919724b69a583a37?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Code Like A Girl","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Code Like A Girl","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/82e21f28c59c499f9c47833c55500fd8","url":"https://daily.dev/sources/colkgirl"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/the-programming-languages-ai-still-gets-surprisingly-wrong-lbxoqyzy8","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":4},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"python,rust","timeRequired":"PT14M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Code Like A Girl","item":"https://daily.dev/sources/colkgirl"},{"@type":"ListItem","position":3,"name":"The Programming Languages AI Still Gets Surprisingly Wrong"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/the-programming-languages-ai-still-gets-surprisingly-wrong-lbxoqyzy8","comment":[{"@type":"Comment","text":"Rust makes sense when you put AI agent into a loop to fix it’s own errors. Python is the interesting one, I thought Go would score higher for being compiled.","datePublished":"2026-08-31T17:29:43.522Z","url":"https://daily.dev/posts/lBXOqYZY8#c-Pr4YH6j7o","author":{"@type":"Person","name":"Isaac de Andrade","url":"https://daily.dev/andradei","image":"https://avatars.githubusercontent.com/u/2653546?v=4"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/the-programming-languages-ai-still-gets-surprisingly-wrong-lbxoqyzy8#faq","mainEntity":[{"@type":"Question","name":"Why does Rust seem harder for AI coding assistants than Python even though the correctness gap isn't that large?","acceptedAnswer":{"@type":"Answer","text":"The perceived difficulty largely comes from the Rust compiler being an unforgiving second reviewer that rejects code immediately, making errors more visible, not from a dramatically higher error rate. A HackerNoon comparative study testing models on 100 problems from October 2025 to February 2026 found the Python-to-Rust performance gap wasn't statistically significant, while Elixir showed a much larger, clearly significant gap. Anyone weighing Rust against other languages for AI-assisted work can track benchmark writeups like this on daily.dev."}},{"@type":"Question","name":"How much does an AI model's knowledge cutoff affect its accuracy with newer Rust APIs?","acceptedAnswer":{"@type":"Answer","text":"Models averaged 56.1% success on Rust APIs available before their training cutoff versus only 32.5% on APIs introduced after, according to the RustEvo\\u00b2 benchmark, which evaluated 588 API changes across the Rust standard library and third-party crates. Giving the model retrieval access to current documentation improved performance on newer APIs by an average of 13.5%. Developers tracking fast-moving crate APIs can follow benchmark research like RustEvo\\u00b2 on daily.dev."}},{"@type":"Question","name":"What's the difference between SWE-bench Verified and SWE-bench Pro scores for AI coding models?","acceptedAnswer":{"@type":"Answer","text":"SWE-bench Verified scores climbed into the 70-80% range for top models by early 2026, but SWE-bench Pro, a harder variant designed to resist memorization and pattern-matching, shows even frontier models scoring closer to 23%. The gap illustrates that benchmark difficulty and realism matter enormously when interpreting a single headline pass-rate number. Anyone comparing coding model claims can weigh benchmark nuances like this via daily.dev before trusting a headline score."}}]} ```