{"version":1,"type":"story","url":"https://digestai.news/story/killswitch-language-challenges-llms-claude-opus-scores-highest","json":"https://digestai.news/story/killswitch-language-challenges-llms-claude-opus-scores-highest.json","markdown":"https://digestai.news/story/killswitch-language-challenges-llms-claude-opus-scores-highest.md","slug":"killswitch-language-challenges-llms-claude-opus-scores-highest","headline":"KillSwitch language challenges LLMs, Claude Opus scores highest","summary":"A developer created KillSwitch, an esoteric stack‑based programming language designed to be hard for large language models (LLMs) and coding agents. The language embeds hidden falsehoods, requires counting letters in identifiers, and includes phrases that trigger model safety guardrails. Most script text is ignored by the interpreter, and literals can be referenced by line numbers, making the syntax unfamiliar to LLMs while remaining usable for humans. The interpreter and specification are open‑source on GitHub, and a full \"Hello World\" example is hosted at killswitch‑lang.org.\n\nThe author benchmarked several major LLMs and coding agents on tasks such as implementing a hello‑world program, explaining script output, and modifying temperature conversion code. Claude Opus 5 achieved the top score of 66.8, outperforming other proprietary models. Open‑weight MiMo‑V2.6‑Pro was the best non‑proprietary model. Models like Fable 5.1 and Gemini 4 Argon performed poorly, reportedly because their safety guardrails are overly sensitive. The benchmark results are posted at bench.killswitch‑lang.org, and the author plans to expand the task set for future model releases.","keyPoints":["KillSwitch is a stack‑based esoteric language built to exploit LLM weaknesses such as letter counting and safety‑guard triggers.","Claude Opus 5 topped the benchmark with a score of 66.8, while open‑weight MiMo‑V2.6‑Pro was the best non‑proprietary model.","Fable 5.1 and Gemini 4 Argon performed poorly, reportedly due to overly sensitive safety guardrails."],"whyItMatters":"The benchmark shows that leading LLMs can still struggle with novel, adversarial code, highlighting limits of current code‑generation claims and prompting more robust evaluation methods.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":[],"models":["Claude Opus 5","Fable 5.1","Gemini 4 Argon","MiMo-V2.6-Pro"],"people":[]},"firstPublishedAt":"2026-10-17T22:26:47Z","updatedAt":"2026-10-17T22:26:47Z","sourceCount":1,"hasPrimarySource":false,"sources":[{"outlet":"blog.picheta.me","title":"I built a programming language that LLMs can't use","url":"https://blog.picheta.me/post/killswitch-lang","publishedAt":"2026-10-17T22:26:47Z","type":"press","primary":false,"lead":true}],"sourceNotes":null,"discussions":[{"site":"Hacker News","url":"https://news.ycombinator.com/item?id=50135762","points":47}],"thread":{"title":"GPT-6 Astra's Emerging Cross‑Domain Conquest","url":"https://digestai.news/thread/stanford-and-caltech-use-gpt-6-astra-in-unitree-g1-robot-to-clean-unfamiliar","storyCount":6},"cite":{"text":"Digest AI, \"KillSwitch language challenges LLMs, Claude Opus scores highest\", 17 October 2026, https://digestai.news/story/killswitch-language-challenges-llms-claude-opus-scores-highest","publisher":"Digest AI","title":"KillSwitch language challenges LLMs, Claude Opus scores highest","datePublished":"2026-10-17T22:26:47Z","url":"https://digestai.news/story/killswitch-language-challenges-llms-claude-opus-scores-highest"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}