**Gemini 3 Pro & Qwen 2.5 VL 72B Shatter AI Benchmarks as Sora 2 Powers Weekly Anime Series**
New AI Models and Benchmark Achievements
Gemini 3 Pro Image (Nano Banana Pro) Dominates LMArena Leaderboard: Google’s latest AI image generation model, Gemini 3 Pro Image (Nano Banana Pro), has set a new record with an 84-point lead on the LMArena leaderboard, showcasing major improvements in prompt adherence and performance. Higher-resolution versions are expected soon.
Qwen 2.5 VL 72B Surpasses Gemini 3 Pro on SpatialBench: Alibaba’s Qwen 2.5 VL 72B has achieved state-of-the-art performance on the SpatialBench benchmark, outperforming Google’s Gemini 3 Pro in spatial reasoning tasks for vision-language models.
AI Image and Video Generation Tools
Qwen-image-edit-2511 Announced for Next Week: ModelScope’s upcoming Qwen-image-edit-2511 focuses on improving character consistency in AI-driven image editing, with a release scheduled for next week.
Sora 2 Used for Weekly Anime Series Production: A creator showcases the use of Sora 2 and Gemini 3 Pro for generating high-quality anime episodes, highlighting cost-effective AI video generation workflows.
AI for Optical Character Recognition (OCR)
Qwen3-2B-VL Excels in OCR Tasks: The Qwen3-2B-VL model demonstrates exceptional OCR performance, with a Dockerized setup available for easy deployment. Its accuracy and efficiency make it a standout in text recognition from images.
AI Development and Evaluation Platforms
Exeta: High-Performance LLM Evaluation Platform: Exeta is a Rust-based platform designed for automated, rigorous testing of LLM-powered features, offering multi-tenant SaaS architecture and production-ready metrics.
Mimir Updates with OAuth and GDPR Compliance: Mimir introduces OAuth support, GDPR++ compliance, and enhanced observability, improving security and reliability for production environments.