--- format: "aidr-story-markdown/v1" id: "18d5dc8473e3848af8fa25e62c11d4366490889bc32fc5d19b10588485429437" canonical_url: "https://aidr.today/18d5dc84?lang=en" title: "Independent Tester's 8.6% Score Contradicts Cantina's 66.7% AI Claim" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-10-06T21:54:08.000Z" category: "Models" topics: ["llm"] source_urls: ["https://huggingnews.com/ai/update-independent-testers-86percent-score-contradicts-cantinas-667perce-84e74586","https://marketbrief.now/ai/update-independent-testers-86percent-score-contradicts-cantinas-667perce-84e74586"] summary: "An independent tester recorded an 8.6% pass rate for Cantina Security's apex-flash-1 model on the CyberGym benchmark, contradicting the developer's claimed 66.7% pass rate. The evaluation found issues including infinite retry loops, tool-call confusion, and context decoherence, and noted that apex-flash-1 is a post-trained version of glm-5.3-flash released as a small run to refine its training process." --- # Independent Tester's 8\.6% Score Contradicts Cantina's 66\.7% AI Claim > [Open the canonical story]() **Published:** 2026-10-06T21:54:08.000Z **Category:** Models **Topics:** llm ## Summary An independent tester recorded an 8\.6% pass rate for Cantina Security's apex\-flash\-1 model on the CyberGym benchmark, contradicting the developer's claimed 66\.7% pass rate\. The evaluation found issues including infinite retry loops, tool\-call confusion, and context decoherence, and noted that apex\-flash\-1 is a post\-trained version of glm\-5\.3\-flash released as a small run to refine its training process\. ## Sources - [Story source]() - [Story source]()