The Smartest Neural Network of 2026: Claude Fable 5 Leads, but GPT-5.6 Sol Is the Best Choice for Most

Independent platform Artificial Analysis has compiled a ranking of neural networks based on nine benchmarks, including scientific code writing, command-line operations, and long document analysis. As of July 22, 2026, the first place was taken by Anthropic's Claude Fable 5 model with a score of 60 points. It also leads on the Text Arena platform, where users blindly evaluate the quality of responses. The key feature of Claude is its ability to handle complex multi-step tasks and analyze dozens of documents, but the cost of processing one test task is about $2.75. Second place with 59 points is held by OpenAI's GPT-5.6 Sol Max, released on July 9, 2026. This model performs the same tasks significantly cheaper—approximately $1.04. Thanks to built-in tools (internet search, program execution, image generation, spreadsheet work), it is considered the best all-round choice for ordinary users. Third place (57 points) belongs to Chinese Kimi K3 from Moonshot AI, which leads in recent tests for creating websites and web interfaces. Fourth is Google's Gemini 3.6 Flash, scoring about 50 points but generating 304 tokens per second (versus 60–72 for the leaders), making it ideal for tasks where speed matters more than accuracy, such as mass file analysis or working within the Google ecosystem. It is noted that the difference between the leaders is only a few percentage points, and technology is advancing so rapidly that the ranking could change within weeks.