OpenAI launches GPT-6.1 Sol, a mid-range model with performance close to Astra and a lower price

OpenAI unveiled the GPT-6.1 Sol at its annual DevDay conference on Tuesday, September 29, in San Francisco. The new model sits in the mid-range position of the GPT-6 family, between the flagship Astra model, launched on September 3, and the Luna, which is designed for simple, high-volume tasks.

Performance and price

The GPT-6.1 Sol offers performance comparable to the GPT-6 Astra, but at around one-fifth of the cost. According to the company, the model is priced at $2 per million input tokens and $10 per million output tokens, compared to $10 and $50 for the Astra, respectively. The previous version, released on September 22, already offered prices half those of the GPT-5.6 generation.

Ultra-fast speed

The Ultrafast mode, now available for both the GPT-6.1 Sol and GPT-6 Astra in the API, ChatGPT Work, and Codex, increases token generation speed by up to eight times in Codex, reaching 350 tokens per second. The service is charged at six times the standard API rate. The feature was first introduced on August 13 in a limited preview, running on GPT-5.6 Sol on Cerebras chips and achieving 750 tokens per second. The partnership with Cerebras aims to deliver 750 megawatts of installed capacity by 2028.

Impact on the portfolio

OpenAI has also canceled the planned October launch of the GPT-6.1 Astra after safety tests revealed lower performance than the current model. As a result, the flagship model remains at version 6, while the Sol advances to 6.1. According to the company, the new model improves performance in programming with agents, computer usage, and professional tasks.

This update reinforces OpenAI’s strategy of providing affordable options for developers who need high performance without the cost of the most advanced model, while also expanding ultra-fast processing infrastructure for real-time applications such as incident response and financial market analysis.