In early September, major AI labs released a series of new models: Anthropic launched Claude Fable 5.1 and Mythos 5.1, Google introduced Gemini 3.8 Flash, Meta debuted Muse Spark 1.3, and OpenAI unveiled GPT-6 Astra. Despite the high frequency of releases, market interest has shifted from benchmark records toward cost-efficiency and practical business application.
OpenAI’s Astra, trained on over 100,000 GPUs at the Stargate facility, costs $10 per million input tokens and $50 per million output tokens. Anthropic’s Fable 5.1 shares similar pricing but offers a 25% to 45% cost reduction for cache reading. Meanwhile, Google’s Gemini 3.8 Flash is priced at $0.75 and $3.75 under a promotion valid until December 31, 2026.
Corporate adoption of top-tier models is slowing as businesses prioritize cost-effective solutions. Anthropic’s Fable 5 accounted for 6% of total token volume, while its more affordable Opus and Sonnet models saw higher usage. OpenAI’s revenue reached over $40 billion annually, driven by the release of the cost-efficient GPT-5.6 Sol.
Anthropic is reportedly preparing for an IPO, aiming to raise up to $100 billion at a valuation of approximately $2 trillion. Nvidia is considered a potential anchor investor with a possible stake of up to $10 billion. Meanwhile, perpetual contracts for Anthropic shares have appeared on crypto exchanges like Binance, Bybit, and Bitget, despite the company declaring such instruments unauthorized.
Analysts estimate the arrival of AGI between 2027 and 2044, with industry leaders narrowing the window to 2026–2030. While models like Astra have achieved high scores on the ARC-AGI-3 benchmark, the industry remains focused on balancing these technical capabilities with the practical financial constraints of enterprise clients.
