Co-Founder & Lead Programmer of AcceleratedLogic AI
GPT-6.1 Sol Pro appears as a separate OpenRouter entry, but OpenRouter describes it as the same underlying GPT-6.1 Sol model served in pro reasoning mode. Its presence beside the standard route represents another execution option. The important difference for a caller is the amount of model work performed per request.
Verified route specifications
OpenRouter route, checked October 3, 2026
Value
Model ID
openai/gpt-6.1-sol-pro
Listing date
September 29, 2026
Context
1,050,000 tokens
Inputs
file, image, text
Output
text
Input price
$2 / 1M tokens
Output price
$10 / 1M tokens
The two Sol routes currently list the same base input and output token rates. OpenRouter cautions that pro mode can consume many more reasoning tokens and take longer to finish, making a typical request more expensive. OpenRouter’s Sol Pro listing
Mode and effort are separate controls
In OpenAI’s Responses API, reasoning.mode selects standard or pro execution. Reasoning effort separately controls depth within that mode. GPT-6.1 Sol defaults to medium effort. To request Pro through the native API, select gpt-6.1-sol and set reasoning.mode to pro; the OpenRouter route name is a provider-specific convenience. Official reasoning-mode documentation
This distinction makes a comparison more useful. A standard/high request and a pro/medium request are different configurations, so an evaluation should retain both settings with the result. Merely writing “tested Sol” would hide the execution mode and leave the cost difficult to interpret.
A cost example
Consider a hypothetical job with 50,000 input tokens. At $2/M, its uncached input costs $0.10. If standard execution uses 5,000 billed output tokens, output adds $0.05; if a deeper execution uses 25,000, output adds $0.25. The total changes from $0.15 to $0.35 even though the per-token price is identical. Those token totals are illustrative, not measurements of Sol Pro.
The decision to use pro should therefore be made per workload. Record completed-task quality, retries, full billed output, and end-to-end latency. Tasks with inexpensive correctness checks can be evaluated by the checks; open-ended analysis needs an explicit review rubric and evidence requirements.
Testing scope
This is a paid route, so it was not submitted to the free-model HTML benchmark in this pass. The article covers configuration and cost mechanics without presenting a new model ranking or a tested accuracy gain. For the standard route’s integration details, see our GPT-6.1 Sol guide.