Smartest Neural Network of 2026: Claude Fable 5 Leads, but GPT-5.6 Sol Is the Best Choice for Most
Anthropic
OpenAI
Moonshot AI
Google/DeepMind
As of July 22, 2026, Anthropic's Claude Fable 5 tops the aggregate benchmark with 60 points, making it the smartest model by raw power. However, OpenAI's GPT-5.6 Sol Max, released on July 9, 2026, scores 59 points while costing significantly less ($1.04 per task vs. $2.75), offering built-in tools for internet search, coding, and image generation. Other notable models include Moonshot AI's Kimi K3 (57 points, excels in web development) and Google's Gemini 3.6 Flash (50 points, but 304 tokens per second for speed-critical tasks). No single model is ideal for all tasks; experts recommend choosing based on specific needs.
As of July 22, 2026, Anthropic's Claude Fable 5 ranks first in an aggregate benchmark from Artificial Analysis, scoring 60 points and outperforming all competitors by a narrow margin. It excels at multi-step reasoning, handling dozens of documents, finding contradictions, and creating coherent plans, but costs about $2.75 per task. OpenAI's GPT-5.6 Sol Max, released July 9, 2026, scores 59 points and costs just $1.04 per task, offering built-in tools for web search, code execution, image generation, and spreadsheet work, making it the best universal choice for most users. Moonshot AI's Kimi K3 scores 57 points and leads in website and web interface creation based on user tests. Google's Gemini 3.6 Flash scores around 50 points in intelligence tests but generates 304 tokens per second, ideal for speed-critical tasks like batch file analysis or working within Google's ecosystem; it also includes the Nano Banana image generator. Experts note that differences between leaders are only a few percentage points, and rapid developments could reshape the ranking within weeks.
Source: Hi-News.ru —
original