GPT-6.1 Sol: OpenAI Brings Astra-Level AI to Codex

Core News and Key Facts

OpenAI’s latest model is making the economics of high-end AI almost as important as the intelligence itself. The company has launched GPT-6.1 Sol, a successor to GPT-6 Sol aimed at coding, computer use and professional workflows. OpenAI says the model delivers near-GPT-6 Astra performance for these tasks at a substantially lower cost. It is available through ChatGPT Work and Codex, while developers can use it through the API.

The API price is $2 per million input tokens, around ₹192 at the exchange rate cited in the supplied report, and $10 per million output tokens, or roughly ₹960. Cached input costs $0.10 per million tokens. OpenAI's current developer documentation confirms these rates and lists a 1.05-million-token context window with up to 128,000 output tokens.

Context and Official Statements

OpenAI positions GPT-6.1 Sol as a lower-cost model for complex coding and professional work. The company says it nearly matches GPT-6 Astra in agentic coding, computer use and professional workflows. Its published model documentation similarly describes GPT-6.1 Sol as offering “near-Astra performance” at lower cost.

The price difference is substantial when compared with Astra. OpenAI's comparison page lists GPT-6 Astra at $10 per million input tokens and $50 per million output tokens, versus $2 and $10 respectively for GPT-6.1 Sol. Cached input is also $0.10 for Sol compared with $1 for Astra.

OpenAI also says an Ultrafast option is coming, with the company targeting much faster token generation. The developer community announcement says Ultrafast can deliver up to 8x faster token generation in Codex and up to 6x in the API.

Krihaa Analysis

The interesting part of GPT-6.1 Sol is not simply that OpenAI has released another model only days after GPT-6 Sol. It shows a change in how frontier AI may be sold: capability is increasingly being separated into different price and speed tiers instead of forcing every developer to pay for the most powerful model.

For Indian startups and software teams, that matters more than benchmark percentages. A coding agent that runs thousands of model calls can become expensive if every request uses a premium model. A near-frontier model at $2 input and $10 output per million tokens gives developers more room to build long-running coding, research and automation workflows.

It also changes the competitive equation for AI applications. If expensive intelligence becomes cheaper, the bottleneck shifts from access to the model toward product design, data, infrastructure and the ability to turn model output into reliable actions.

The upcoming Ultrafast tier adds another layer: developers may soon choose between intelligence, cost and latency rather than simply choosing between different AI models. For Hyderabad’s software ecosystem, that could make sophisticated AI agents more practical for smaller teams because inference costs become easier to fit into real product economics.

The broader direction is clear: frontier AI is moving from a “best model” race toward an infrastructure market where price per task, response speed and reliability can matter as much as benchmark scores.