--- format: "aidr-story-markdown/v1" id: "f4e6102360faadd21ca90d639146e15c920805627db3d0c10c567272e9d4f0c0" canonical_url: "https://aidr.today/f4e61023?lang=en" title: "Google Gemini 4's Coding Skills Trail Benchmarks in Internal Tests" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-09-30T20:23:26.000Z" category: "Models" topics: ["google","coding"] source_urls: ["https://marketbrief.now/ai/google-gemini-4s-coding-skills-trail-benchmarks-in-internal-tests-5b2d6bff","https://huggingnews.com/ai/google-gemini-4s-coding-skills-trail-benchmarks-in-internal-tests-5b2d6bff"] summary: "Personnel with direct access to the company's next-generation artificial intelligence model report it frequently fails at programming duties despite high industry efficacy scores. Testers noted that Gemini 4 struggles with front-end design and other coding tasks, and warned that the model's large size could make it expensive to operate. Google disputes these claims and states there is broad internal consensus that the model remains at the frontier of AI development. The company abandoned plans to launch Gemini 3.5 Pro in June to prioritize the debut of Gemini 4. While some staff believe the new model still lags behind current offerings from OpenAI and Anthropic in specific areas, others argue it has caught up. The flagship model is reported to perform particularly well in cybersecurity, natural communication, and multimodal tasks including video understanding." --- # Google Gemini 4's Coding Skills Trail Benchmarks in Internal Tests > [Open the canonical story]() **Published:** 2026-09-30T20:23:26.000Z **Category:** Models **Topics:** google, coding ## Summary Personnel with direct access to the company's next\-generation artificial intelligence model report it frequently fails at programming duties despite high industry efficacy scores\. Testers noted that Gemini 4 struggles with front\-end design and other coding tasks, and warned that the model's large size could make it expensive to operate\. Google disputes these claims and states there is broad internal consensus that the model remains at the frontier of AI development\. The company abandoned plans to launch Gemini 3\.5 Pro in June to prioritize the debut of Gemini 4\. While some staff believe the new model still lags behind current offerings from OpenAI and Anthropic in specific areas, others argue it has caught up\. The flagship model is reported to perform particularly well in cybersecurity, natural communication, and multimodal tasks including video understanding\. ## Sources - [Story source]() - [Story source]()