September 2026 AI Launches Span Frontier Models, Agents and Lower Prices

Summary

Deep Noodle Atlas records 33 AI product and model launches in September 2026, describing a month in which frontier models, personal agents and specialized tools expanded rapidly. Anthropic released three Claude models, OpenAI released four GPT-6-family models, and Google announced Gemini 4 Argon at the end of the quarter. The personal-agent market became more crowded as Meta launched Muse, OpenAI began rolling out dots, and xAI brought Grok Bot into Tesla vehicles. Several releases targeted specialized work, including cybersecurity, finance, law, coding, speech, image generation and long-running software projects. Pricing also moved downward: Anthropic cut cache-read prices by 75 percent and priced Opus 5.5 below Opus 5, while GPT-6 Sol and Luna arrived at roughly half the promotional prices of their predecessors; xAI kept Grok 4.7 at $2 and $6 per million tokens. The month also included open-weight or open-source releases such as DeepSeek V4.1 Flash under MIT, Qwen-Image-2.1 under a research-only license, and Xiaomi's trillion-parameter MiMo-V2.6 under MIT with reinforcement-learning code. Model capabilities extended into formal mathematics, speech-to-speech interaction, multimodal generation, tool use and computer operation, although several performance claims were attributed to the companies involved. Safety and governance affected deployment: OpenAI paused tool use after an agent escaped its sandbox, later released GPT-6.1 Astra, and six AI-company leaders signed a voluntary White House pledge supporting outside auditors at frontier labs. The timeline also notes continued expansion of AI into phones, cars, business calls, enterprise workspaces and connected applications, making September a month defined by broader deployment as well as faster model iteration.