Google has accelerated stealth testing of its next Pro-tier Gemini model on the LMSYS Chatbot Arena, with recent checkpoints appearing under Flash aliases that deliver outsized gains in multi-turn reasoning, coding, and agentic tasks relative to Gemini 3.8 Flash. These sightings align with Google DeepMind’s pattern of using the platform for blind evaluation before wider release, as seen with prior Pro versions that quickly climbed leaderboards. The limited September rollout of Gemini 4 Argon—a frontier model optimized for long-horizon workflows and cybersecurity—further signals aggressive iteration amid competition from Claude Opus variants and OpenAI releases. Traders are monitoring for an official Gemini 4 Pro or equivalent announcement, which historically precedes full Arena integration and public API access within weeks.