Microsoft launched the preview version of MAI-Image-2.7, ranking second globally on the Artificial Analysis leaderboard behind OpenAI. The model specializes in instruction-based image editing (style transfer, relighting, text modification) but currently lacks text-to-image capabilities and public API access, available only via Playground.
Elo Rating System
Elo评分系统
A method for calculating relative skill levels based on win probabilities, commonly used in model blind test rankings.
Curated Intelligence
Artificial Analysis introduced Code2Video-AA, an independent benchmark evaluating how well models generate marketing videos via code. Claude Opus 5.5 leads the rankings but costs ~10x more than GPT-6.1 Sol. The benchmark assesses performance across five axes using 114 design briefs and over 35,000 pairwise comparisons to calculate Elo ratings.
Artificial Analysis launched Code2Video-AA, an independent implementation of HeyGen's Code2Video Bench, evaluating how well LLMs design short videos using HyperFrames HTML code. Claude Opus 5.5 leads with a significant Elo score advantage, ranking first across all five judge axes. This benchmark provides developers with quantitative insights into model performance in creative coding and multimodal output scenarios.
Moonshot AI released Kimi K3.1 Pro, which ranks among leading models in agentic knowledge work and coding benchmarks like GDPval-AA and Terminal-Bench 4.0. While early signs suggest it may rank lower than its predecessor on the AA-3D benchmark for interactive 3D scene building, a demo of a complex mountain-bike descent highlights its potential capabilities.
HeyGen released HeyGen Voice Turbo, which secured the #1 spot on the Artificial Analysis Controlled Voice Arena Leaderboard. The model offers significant cost efficiency at $4/1M characters and demonstrates improved pronunciation robustness and accent handling compared to its predecessor and competitors.
Artificial Analysis introduces AA-PersonalityBench, evaluating LLMs across 30 personality traits and 21 communication characteristics using 900 curated questions. The benchmark aggregates subjective traits into eight indexes (e.g., Expressiveness, Transparency) and objectively quantifies style metrics like punctuation and emoji usage, aiding users in selecting models suited for specific contexts such as creative writing or legal tasks.
Arena's Week 42 blind test rankings show stable leadership from Gemini and Claude models. Chinese LLMs have solidified their mid-tier positions across domains, with Kimi K3 Max entering the top 10 for coding and Xiaomi's Mimo V2.6 Pro reaching the top 20 in front-end development. The newly listed Gemini 3.5 Flash Medium immediately entered the top 30 overall.
Black Forest Labs released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model that outperforms closed competitors like Ideogram V4.5 in internal Elo tests. It supports native 4K output, pixel-accurate editing, and multi-reference generation while running on consumer GPUs with just 8 GB of VRAM.
Artificial Analysis introduces AA-3D, a new benchmark evaluating how well AI models act as agents to build interactive three.js scenes from text briefs in an offline sandbox. Ranked by over 10,000 blind human votes using the Elo system, results show four models tied at the top led by Claude Sonnet 5.5, while the cheapest model, GPT-6 Luna, ranks eighth, highlighting significant disparities in spatial logic and code generation capabilities.
Arena released its Week 41 LLM leaderboard for 2026, where Anthropic's newly launched Claude Opus 5.5 Max immediately claimed the top spot in the Agent category upon debut. In the overall rankings, Gemini-4-Argon-High maintained its lead, while Chinese models like Kimi-K3-Max and Qwen3.8-Max held stable positions in the mid-upper tier. The ranking reflects user subjective preferences via blind voting and ELO scores rather than absolute capability benchmarks.