Artificial Analysis's release page lists six GPT-6 Sol configurations. The Intelligence Index v4.3.2 ranges from 28 for Non-reasoning to 48 for max; output speed ranges from 109 to 129 tokens/s; and the weighted average cost per Intelligence Index task ranges from $0.13 to $1.06. The page gives time to first answer token only for Non-reasoning (0.93 seconds); latency for the other configurations is not listed.
Intelligence Index: Artificial Analysis Intelligence Index v4.3.2. The page says it includes 10 evaluations: AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience, and AA-LCR v1.1.
Cost: The chart defines this as the weighted average dollar cost per Intelligence Index task. The breakdown chart says costs are split by token type and include Answer, Reasoning, Cache Write, Cache Hit, and Input. Total task costs for each configuration are shown in the table.
Speed: Output Speed, measured in tokens/s; the page labels higher as better.
Latency: The page summary defines latency as time to first answer token; it lists only 0.93 seconds for Non-reasoning.
Not disclosed: The page does not specify per-evaluation scores, per-question token counts and durations, sample size, number of runs, prompts, API provider, region, hardware, or concurrency settings.
| GPT-6 Sol configuration | Intelligence Index v4.3.2 (score) | Output speed (tokens/s) | Intelligence Index cost per task (USD/task) | Time to first answer token (seconds) |
|---|---|---|---|---|
| max | 48 | 115 | 1.06 | Not listed |
| xhigh | 44 | 128 | 0.53 | Not listed |
| high | 43 | 119 | 0.37 | Not listed |
| medium | 40 | 114 | 0.25 | Not listed |
| low | 34 | 129 | 0.13 | Not listed |
| Non-reasoning | 28 | 109 | 0.33 | 0.93 |
The page notes that max has the highest Intelligence Index among the six configurations, while low has the highest output speed and the lowest task cost. Non-reasoning has the shortest time to first answer token. The speed ranking does not match the intelligence or cost rankings.
The scores, speed, task cost, and latency in the table are values displayed directly on the page; the page does not publish the full calculation method or confidence intervals for these aggregate values.
Task cost is a weighted average estimate for the AA Intelligence Index task set. It is not the bill for a fixed business request, nor an input or output token price. Do not confuse it with the $1.5 Price field in the page's Further details table; that field does not state a unit or calculation method in the same table.
Speed is an output-speed metric in tokens/s. The page does not explain the prompt length, output length, or statistical method used for measurement, so this value cannot be used to infer end-to-end completion time for a specific task.
The page does not provide first-token latency for the other five configurations beyond Non-reasoning; latency cannot be inferred from output speed.
An existing Artificial Analysis article, "GPT-6 Sol and Luna Push the Cost-Efficiency Frontier," records an Intelligence Index of 48 and a task cost of $1.06 for GPT-6 Sol max, matching two figures on this page. The article labels the index version v4.3. This page adds horizontal results for the other five configurations, output speeds for all six, and the release page's own v4.3.2 version label. The version labels from the two pages are recorded separately and are not combined to infer anything.
GPT-6 Sol
