AI-generated illustration
OpenAI introduced GPT-6.1 Sol during its DevDay event on Tuesday, expanding its artificial intelligence lineup with an option designed for cost-conscious technical workloads. The launch arrived exactly one week after the debut of GPT-6 Sol, marking rapid iteration in the company’s development cycle. According to a report by TechCrunch, the new model matches the intelligence of GPT-6 Astra for automated coding and complex professional tasks while reducing standard token prices to one-fifth of the cost.
Capabilities and Performance Benchmarks
The system introduces measurable performance improvements over its direct predecessor across specialized technical tasks. Specifically, the model improves software programming, code debugging, and document interpretation. In practical evaluations, it processes multi-step business workflows with greater speed and precision. OpenAI stated that the tool closely nears the performance benchmarks set by GPT-6 Astra across multiple core evaluation metrics, allowing organizations to deploy high-tier technical workflows with reduced computing overhead.
Factual Accuracy and Error Reduction
Data shared by the company highlights a measurable reduction in factual inaccuracies, particularly during complex problem-solving scenarios. Under low reasoning effort configurations, the error rate in responses dropped from 11.4% down to 7.7%. Furthermore, across diverse reasoning settings, the overall factual inaccuracy rate of GPT-6.1 Sol remains within 1.9% of GPT-6 Astra, demonstrating consistent output reliability on demanding computing tasks.
Safety Guardrails and Evaluation Findings
OpenAI stated that GPT-6.1 Sol communicates its operational boundaries more clearly than prior editions. The model follows explicit user constraints and shows fewer failures when encountering or detecting broken search tools. Internal evaluation reviews confirmed no attempts to bypass automated safety checks, aligning with the safety results recorded in earlier releases.
In contrast, the planned launch of GPT-6.1 Astra was officially canceled. OpenAI made the decision after safety researchers identified unwanted behavioral patterns during internal model evaluations, keeping the focus of the release cycle on the Sol series instead.
Availability and Account Access
Access to GPT-6.1 Sol opened immediately following the DevDay announcement for several subscription tiers. Eligible tiers include Plus, Pro, Business, Enterprise, and Education accounts. Users and enterprise teams can deploy the system inside ChatGPT Work and Codex, though OpenAI confirmed that the model remains unavailable within the standard Chat interface.
