OpenAI has announced a new model tier: GPT-6.1 Sol. The headline for businesses running AI automations is a massive drop in API costs for high-level capabilities.
According to the announcement, GPT-6.1 Sol offers "near-Astra intelligence" specifically tuned for coding, computer use, and professional work. The standout detail is the price: it costs exactly one-fifth of Astra’s standard API input and output token prices.
At this stage, official details remain strictly limited to this core announcement. We do not yet have extensive performance benchmarks or availability dates, but the initial pitch is clear: frontier-level reasoning at an 80% discount.
What This Means for SMB Automation
For small and mid-sized businesses, API token costs are the primary recurring expense in any AI automation deployment. When a model processes complex tasks—like reading through a messy database, categorizing inbound customer requests, or navigating software via "computer use" features—it consumes a massive amount of tokens.
Until now, businesses often had to choose between highly capable, expensive models and cheaper, less reliable ones. If GPT-6.1 Sol delivers on its promise of near-Astra intelligence for professional work at a fraction of the cost, it fundamentally changes the return on investment for operational automation.
An 80% reduction in API costs means:
- More complex workflows become viable: Multi-step workflows that require thousands of tokens per task won't blow out a software budget.
- Scalability: Processing higher volumes of daily admin tasks suddenly makes more financial sense.
- Better baseline performance: Businesses currently using smaller, cheaper models to save money might be able to upgrade their system's reasoning capabilities without increasing their monthly spend.
Our team at Install Agent builds these types of administrative and operational automations every day, and we know that in the real world, "near-Astra" remains an unverified claim until tested. The true benchmark will be how GPT-6.1 Sol handles unpredictable edge cases in daily operations compared to its more expensive siblings. We will be watching closely as more details emerge and the model becomes available for hands-on evaluation.