Arena

MAI-Image-2.5 Ranks Differ Across Arena Snapshots

Published evidence shows MAI-Image-2.5 was reported at No. 2 on Arena's image-editing leaderboard in Microsoft AI launch material, and No. 3 for text-to-image, while an Arena single-image editing snapshot captured on August 7, 2026 placed MAI-Image-2.5 at No. 4 (with Muse Image at No. 3 and GPT-Image-2 at No. 1). The article highlights that benchmark claims require explicit task categories, dated sources, and defined comparator sets; the supplied public snapshots lack sufficient methodological detail to evaluate Arena's evaluation design. For business procurement of generative image tools, the piece recommends documenting category, snapshot date/source, comparison set, workflow testing, and governance evidence rather than relying on an undated or category-mismatched rank.

Google releases Gemini 3.7 Flash for coding and agents

Google released Gemini 3.7 Flash on 2026-08-14, a workhorse LLM optimized for coding, web development and multi-step agent workflows. Google says 3.7 Flash improves first-try code quality, long-task stability, instruction-following, multi-step planning, tool use and safety protections, and is integrated across the Gemini API, Google AI Studio, Android Studio, enterprise offerings and Gemini Spark for Google AI Pro/Ultra. Public benchmarks show gains in coding, document understanding and business-process automation versus Gemini 3.6 Flash and competitive performance against GPT-5.6 Terra and Claude Sonnet 5 in several tests. Google halved introductory token pricing versus the previous Flash release to make production AI-agent deployments more cost-effective, and updated safety measures targeting biological, chemical, radiological, nuclear and cyber misuse scenarios.

General Catalyst leads $1.1B into River AI

River AI, a startup founded by xAI co-founder Igor Babuschkin, raised $1.1 billion in a seed/Series A round led by General Catalyst and AMP PBC, with participation from Nvidia, AMD Ventures, Y Combinator, and Temasek. River, which emerged from stealth in June, aims to rebuild the AI stack end-to-end — training, models, product layer, and new hardware — to enable personally trainable AI agents. The company already offers an API billed per 1 million tokens supporting reinforcement learning and LoRA fine-tuning. AMP PBC was founded in 2026 by former Andreessen Horowitz general partner Anjney Midha. River positions its “neocloud” and tooling as enabling enterprises and individuals to train and serve open models with faster RL runs and claimed cost savings versus closed-source alternatives.