Four Frontier AI Models Launched in 30 Days

Anthropic, OpenAI, and Google DeepMind collectively launched four frontier-class AI models within a concentrated 30-day period in September 2026. Claude Fable 5.1 was released first on September 1, followed by OpenAI's GPT-6 Astra on September 3. The month concluded with the release of OpenAI's GPT-6.1 Sol and Google DeepMind's Gemini 4 Argon in the final days of September. While launch announcements often highlight benchmark performance, a closer examination reveals overlapping scores, suggesting that differences in pricing, access policies, and cost per task are more significant differentiators for users and developers.

A notable development occurred on September 28, when OpenAI cancelled the planned release of GPT-6.1 Astra. This decision followed internal scope and authorization tests that the model reportedly failed. Consequently, GPT-6 Astra remains OpenAI's leading model for the time being. The pricing structures for these models vary considerably. GPT-6 Astra and Claude Fable 5.1 share an identical list price of $10 per 1 million input tokens and $50 per 1 million output tokens. In contrast, GPT-6.1 Sol and Gemini 4 Argon are priced at one-fifth of this rate, with Gemini 4 Argon's lower price being an introductory offer that is expected to double later. For cached inputs, which are crucial for AI agents that resend system prompts and historical data with each step, Astra's $1.00 per 1 million tokens is four times the cost of Fable 5.1's and ten times the cost of Sol's. Gemini 4 Argon's introductory cached input price is $0.10 per 1 million tokens.

Beyond standard pricing, long-context pricing also presents a significant divergence. For contexts exceeding 272,000 tokens, GPT-6 Astra is priced at $20 per 1 million input tokens and $75 per 1 million output tokens. GPT-6.1 Sol's long-context pricing is set at double the input rate and 1.5 times the output rate of its short-context tier. Claude Fable 5.1 offers flat pricing up to 1 million tokens, while Gemini 4 Argon's long-context pricing details have not been disclosed. The context window sizes also differ, with GPT-6 Astra and GPT-6.1 Sol both supporting 1.05 million tokens, and Claude Fable 5.1 supporting 1 million tokens. Gemini 4 Argon's context window size is not disclosed. However, Gemini 4 Argon stands out with a maximum output per response of 1 million tokens, significantly larger than the 128,000 tokens offered by the other three models. All four models offer varying levels of reasoning control, with GPT-6 Astra and GPT-6.1 Sol providing five effort levels from low to maximum, and Claude Fable 5.1 including low, medium, and high effort levels. Gemini 4 Argon's reasoning control levels are not disclosed. None of these frontier models are open-weight.