GPT-5.2-high LMArena scores released, OpenAI falls from #6 to #13 (lmarena.ai)

Google’s Gemini 2.5 Pro remains the top performer on the latest public leaderboards, leading across both text and vision arenas in a fresh snapshot that compares models on text, image, vision and other multimodal tasks. The leaderboard page presents per-arena rankings and raw stats (with deeper drilldowns available in each model’s tab), and shows Gemini 2.5 Pro consistently at or near the top of aggregated and specialized evaluations against many competitors across diverse test sets. For the AI/ML community this reinforces that cutting-edge multimodal models are converging on stronger unified capabilities: high-quality language understanding and image/vision reasoning in a single model. Practically, that means better out‑of‑the‑box performance for multimodal applications (image captioning, visual question answering, multimodal search) but also sharper expectations for compute, latency and safety tradeoffs when deploying these large models. The leaderboard snapshot is a useful reference for benchmarking and model selection, while reminding researchers that ongoing evaluation diversity (different arenas, datasets and metrics) remains crucial as vendors iterate and rivals close the gap.