OpenAI delivers high-end intelligence at low-end costs with GPT-6.1 Sol

OpenAI has pulled back the curtain on GPT-6.1 Sol, a refreshed iteration of its core artificial intelligence model designed to bring flag-ship capabilities to developer workflows without breaking the bank. The newly unveiled framework marks a pivotal shift in the artificial intelligence market, striking a calculated balance between sophisticated computing power and economic efficiency. By providing technical teams with access to near-top-tier operational intelligence at a fraction of the cost, the organization aims to unlock complex automation for businesses that were previously priced out of deploying large-scale autonomous systems.

At the heart of GPT-6.1 Sol is an architecture calibrated to deliver up to eighty percent cost savings across both input and output operations compared to OpenAI’s top-tier flagship model, GPT-6 Astra. Developers building autonomous software agents now face a structured pricing environment set at two dollars per million input tokens and ten dollars per million output tokens. This reduced financial threshold allows research departments and enterprise teams to execute heavy-duty programming, navigate computer desktop operating systems, and handle complex enterprise pipelines at scale without incurring astronomical operating expenses.

A major driver behind these operational savings stems from OpenAI’s aggressive pricing structure for cached input data. Fixed at ten cents per million tokens, the caching framework allows the system to store and repeatedly reference massive files—such as extensive code repositories, complete software documentation, or legal dossiers—without demanding fresh processing cycles for every subsequent prompt. This setup represents a ninety-five percent cost drop relative to standard input rates and effectively halves the caching expenses of the previous GPT-6 Sol generation, freeing up significant infrastructure capacity on backend servers and passing those direct operational savings to end-users.

Performance benchmarks confirm that this dramatic drop in price does not come with severe sacrifices in raw capability. In software engineering evaluations, particularly on the DeepSWE v1.1 benchmark designed to test models against complex real-world codebases, GPT-6.1 Sol matches the success rates of GPT-6 Astra while drawing only twenty percent of the budget. It also comfortably outperforms its predecessor by 6.4 percentage points while requiring substantially less reasoning effort, offering engineering teams a cost-effective workhorse for real-time debugging, automated code generation, and repository management.

The model demonstrates equal versatility across complex document processing and enterprise workflow management. Testing on the GDP.pdf benchmark highlights its capacity to process unstructured text, complex statistical tables, and embedded diagrams from technical files with accuracy that rivals Anthropic’s Opus 5.5 at less than half the per-task execution cost. Furthermore, on AutomationBench, which evaluates multi-step business actions across nearly fifty distinct software tools for sales, operational support, and marketing, GPT-6.1 Sol posted higher task completion rates than competing models while maintaining a one-third expense profile.

Beyond document parsing, the upgrade delivers substantial progress in desktop environment navigation and computational science. Benchmarks on OSWorld 2.0 show that GPT-6.1 Sol outperforms the original Sol release by seven percentage points in full desktop control tasks at less than half the execution cost, trailing the flagship Astra by a negligible margin. In scientific computing environments like Terminal-Bench Science 0.1, the new system doubled its performance over prior standard iterations, driving average per-task expenses down from over twenty-three dollars to just under six dollars. OpenAI has integrated GPT-6.1 Sol directly into its Plus, Pro, and Business ecosystem—including ChatGPT Work and Codex—with an Ultrafast high-speed variant expected to join the line-up shortly.