In Artificial Analysis's evaluation, moving Gemini 4 Neon from High to Xhigh raised its overall index score from 60 to 61, added 6.7 minutes per index task, and increased the weighted cost per task by $2.70.
What this evaluation can assess: Differences in overall index score, time to the first answer, and evaluation cost across the four reasoning tiers of the same model.
Limits on extrapolation: Success rates on specific business tasks, latency in chat clients, and actual bills for inputs of different lengths.
Model version: Google Gemini 4 Neon, with gemini-4-neon as the model page slug.
Test environment or client: Artificial Analysis's independent evaluation. The source page's FAQ says task duration and first-answer latency are based on the Google API.
Reasoning tiers and parameters: Low, Medium, High, and Xhigh. The source leaves the exact API endpoint, model snapshot ID, temperature, and other generation parameters unspecified.
Artificial Analysis Intelligence Index v5.0.1 combines nine evaluations: AA-Briefcase v1.1, AutomationBench-AA, Terminal-Bench 4.0, FrontierCode-AA 2.0, AA-Omniscience v1.1, GDP.pdf, HLE-Diamond, CritPt v2 (Hard Subset), and Terminal-Bench-Science 0.1.
The page estimates the duration of an index task by dividing its output token count by the end-to-end output speed. End-to-end output speed is the output token count divided by the full request duration, including the wait for the first token. First-answer latency measures time to the first answer token, including the time spent reasoning before the answer begins.
| Reasoning tier | Overall index score | Index task duration (minutes) | End-to-end output speed (tokens/s) | First-answer latency (seconds) | Weighted cost per index task (USD) |
|---|---|---|---|---|---|
| Low | 44 | 3.7 | 97 | 13.93 | 0.75 |
| Medium | 55 | 8.3 | 101 | 20.14 | 1.90 |
| High | 60 | 14.8 | 101 | 40.64 | 3.74 |
| Xhigh | 61 | 21.5 | 110 | 56.51 | 6.44 |
| Reasoning tier | Weighted output tokens per index task | Cost to run the full index (USD) |
|---|---|---|
| Low | 21,726 | 830 |
| Medium | 50,000 | 1,866 |
| High | 90,188 | 3,695 |
| Xhigh | 141,631 | 7,788 |
The pricing charts show the same rates per million tokens for all four tiers. Cache hits cost $0.15, input costs $3.00, and output costs $15.00. A separate chart on the source page gives the cost to run the full index, which covers all nine evaluations.
Moving from Low to Medium adds 11 index points, the largest gain between adjacent tiers. Xhigh has the highest end-to-end output speed and the highest weighted output token count per task. With that token count, Xhigh still has the longest estimated task duration of the four tiers.
The source page leaves per-tier sample counts, per-question results, the full test-tool configuration, and per-request parameters unspecified for these aggregate figures. The figures reflect the page as it appeared on 2026-11-19. Collect fresh figures if the index version or provider performance changes.
Open the source page and confirm the Intelligence Index version and the four tier names in the Further details table. Hover over the corresponding bars in Cost per Intelligence Index Task, Cost to Run Artificial Analysis Intelligence Index, Time to First Answer Token, Output Tokens per Intelligence Index Task, and Pricing in that order. Check the tables against the chart values in the units given for each metric. A full rerun requires API request settings and sample details that the source page does not publish.
Gemini 4 Neon