Did Google just dethrone OpenAI and Anthropic?
Google's Gemini 4 Argon claims top benchmarks over GPT-6 and Claude Opus 5.5
“it's a sparse mixture of experts transformer and was trained from scratch on Google's TPUs”
Google released Gemini 4 Argon, a new frontier model built from scratch using a sparse mixture-of-experts architecture on Google TPUs, claiming state-of-the-art performance on software engineering and multi-domain benchmarks while beating GPT-6 and Claude Opus 5.5. The model was initially released to a limited 'Fairwind' program without guardrails before broader rollout. This is a Fireship commentary video with satirical framing, so benchmark claims should be treated as directionally informative rather than rigorously sourced.