Artificial Analysis places GPT-6.1 Sol one point below GPT-6 Astra on its Intelligence Index, while estimating that Sol costs $0.72 per indexed task against $3.26 for Astra. That puts Sol at roughly 22% of Astra’s estimated task cost. Put the other way, Astra costs about 4.5 times as much per task in this comparison.
Why it matters. OpenAI said when it launched Sol that it nears Astra’s capability at a fraction of the price, and the evidence at the time was OpenAI’s own. Artificial Analysis is an outside evaluator, so this is the first independent measurement of that claim. It lines up roughly with OpenAI’s one-fifth framing, but it isn’t the same measure: OpenAI’s claim concerned token rates, while Artificial Analysis estimates cost per task on its own benchmark set.
What OpenAI lists. OpenAI introduced GPT-6.1 Sol for coding, computer use, and professional work, with pricing of $2 per million input tokens, $10 per million output tokens, and $0.10 per million cached-input tokens. Artificial Analysis also reports that GPT-6.1 Sol replaced GPT-6 Sol after seven days. Its comparison describes Sol as having near-Astra intelligence, but the available material doesn’t explain the replacement, quantify the improvement over GPT-6 Sol, or document migration effects for existing users.
What the comparison doesn’t show:
Real deployments. The figures come from Artificial Analysis’ benchmark and cost method. They don’t show how the models compare on particular coding workloads, computer-use flows, latency, reliability, tool behavior, or context limits.
How big a point is. The material doesn’t give the index scale, so whether a one-point gap is negligible or meaningful isn’t clear from these sources.
Your own bill. The $0.72 and $3.26 are estimates for indexed tasks. Real cost depends on prompt length, retries, caching, and how often a cheaper model needs a second attempt, the same lesson as the Databricks benchmark.
Bottom line: On Artificial Analysis’ index, Sol is close to Astra at a much lower estimated cost per task. That makes it a strong candidate to test, not a ready-made replacement. Teams should run their own workloads, tracking task success, retries, latency, and total cost, before treating a one-point gap on an aggregate index as a deployment decision.
