OpenAI GPT-6.1 Sol Cuts AI Costs Without Sacrificing Performance: Here's Everything You Need to Know

OpenAI has launched GPT-6.1 Sol, a new AI model designed for coding, complex document analysis and agentic computer-use tasks. While the model is positioned as an upgrade over GPT-6 Sol, its bigger selling point is its cost-to-performance ratio. OpenAI says GPT-6.1 Sol can deliver performance close to its more capable GPT-6 Astra model on several demanding benchmarks while costing a fraction as much.

The launch comes as developers and businesses increasingly use AI agents for tasks that require multiple model calls, tool interactions and reasoning steps. In these workflows, the cost of running an AI model can become just as important as its intelligence. OpenAI is therefore positioning GPT-6.1 Sol as an option for companies that need stronger reasoning than GPT-6 Sol without paying GPT-6 Astra-level prices for every task.

GPT-6.1 Sol Focuses on Coding and Business Workflows

OpenAI says GPT-6.1 Sol improves coding, debugging, document understanding and multi-step business automation compared with GPT-6 Sol.

On DeepSWE v1.1, a benchmark focused on complex software-engineering tasks in real codebases, GPT-6.1 Sol scored 6.4 percentage points higher than GPT-6 Sol while using lower reasoning effort and cost. OpenAI says it roughly matched GPT-6 Astra on the same benchmark at around one-fifth of the cost.

That difference could matter for software agents, which often need to make repeated calls to an AI model while analysing code, using tools and testing solutions. Even a relatively small reduction in the cost of each interaction can become significant when an agent performs thousands of tasks.

GPT-6.1 Sol also performed strongly on GDP.pdf, which evaluates how AI models handle professional PDFs containing tables, charts, diagrams and fine-print information. OpenAI says it outperformed Opus 5.5 with fallbacks at less than half the cost per task and approached GPT-6 Astra's performance at around one-fifth the cost.

On AutomationBench, GPT-6.1 Sol scored 2.2 percentage points higher than Opus 5.5 at medium reasoning effort while costing roughly one-third as much. It also scored 4.8 percentage points higher than GPT-6 Sol under the same conditions.

Computer-use and Science Performance gets a Boost

OpenAI is also highlighting improvements in computer-use workflows. On the offline set of OSWorld 2.0, GPT-6.1 Sol scored seven percentage points higher than GPT-6 Sol at maximum reasoning effort while costing less than half as much.

The model came within 2.1 percentage points of GPT-6 Astra while costing roughly one-seventh as much per task, according to OpenAI.

Scientific workloads also showed substantial gains. On Terminal-Bench Science 0.1, which includes data analysis, simulation and theorem-proving tasks, GPT-6.1 Sol more than doubled GPT-6 Sol's score at maximum reasoning effort while costing less than half as much.

OpenAI says GPT-6 Astra still achieved the highest score in this evaluation at 68.1% and remains aimed at the most difficult scientific research tasks. This suggests GPT-6.1 Sol is not intended to replace the company's highest-end model across every workload, but rather to offer a more economical alternative for many demanding tasks.

GPT-6.1 Sol also Claims Better Factuality

OpenAI says GPT-6.1 Sol reduces factual errors compared with GPT-6 Sol. At low reasoning effort, the share of responses containing a factual error fell from 11.4% with GPT-6 Sol to 7.7% with GPT-6.1 Sol, representing an approximate 32% reduction.

The company says the model's error rate remained within 1.9 percentage points of GPT-6 Astra across tested reasoning settings while costing less than one-fifth as much.

However, OpenAI notes that the factuality evaluation used deliberately difficult, de-identified conversations where users had previously reported errors. As a result, the results should not be interpreted as a direct measure of everyday accuracy.

GPT-6.1 Sol Pricing and Availability

Through the OpenAI API, GPT-6.1 Sol is priced at $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens.

OpenAI says cached input costs 95% less than standard input pricing and 50% less than GPT-6 Sol's cached input pricing. A GPT-6.1 Sol Ultrafast version is also expected in the coming days, with up to eight times faster token generation in Codex.

GPT-6.1 Sol is available from September 30 to Plus, Pro, Business, Enterprise and Edu users through ChatGPT Work and Codex. It is not currently available in Chat.

For developers, the bigger story is the economics. If GPT-6.1 Sol can deliver much of the performance of a flagship model at significantly lower prices, it could make sophisticated AI agents more practical to deploy at scale.