Big Models Fail - Claude Opus 4.6, GPT-5.2 Score Only ~30% on New Coding Text скачать в хорошем качестве

Big Models Fail - Claude Opus 4.6, GPT-5.2 Score Only ~30% on New Coding Text

In today’s episode of Front Page by AIM Network, we’re breaking down the most awkward moment in AI history and the benchmark that just exposed the industry’s biggest secret. While Sam Altman (OpenAI) and Dario Amodei (Anthropic) were busy dodging a handshake at the India AI Impact Summit, a new test called SWE-Atlas by Scale AI was putting their "God models" through the wringer. The result? The hype is hitting a wall. The SWE-Atlas Brutality: Why top models like Claude Opus 4.6 and GPT-5.2 scored as low as 30% on real-world engineering tasks. Systems vs. Snippets: Why AI can write a function but can’t understand a codebase. The Tool Factor: How taking away terminal access drops AI performance by nearly 50%. The Future of the AI Engineer: Who is winning the race to build a model that can actually work? The biggest myth in tech is that coding is "solved." This data proves it isn't. We are moving from the era of "Chatbots" to the era of "AI Agents," and the transition is getting messy. Think AI. Think AIM. ------------------------------------------------------------------------------------------ 👋 ABOUT AIM NETWORK AIM Network is India's leading voice on the intersection of AI, Technology, and Business. We bring you sharp analysis, breaking news, and original perspectives on everything from Generative AI breakthroughs to geopolitical market shifts. 🔔 Subscribe for daily insights:    / @aimmediahouse   📺 WATCH OUR POPULAR SERIES: Front Page (Daily Tech News):    • Front Page   Tech Maccha (Deep Dives):    • Tech Maccha   Tech Talks (Podcast):    • Tech Talks   📰 STAY UPDATED: Website: https://analyticsindiamag.com/ LinkedIn:   / analytics-india-magazine   Twitter/X: https://x.com/Analyticsindiam Instagram:   / aim.media.house   ------------------------------------------------------------------------------------------