October 14, 2026
xhigh in Claude Code: Sonnet 5.5 solves a task 77% cheaper than on max
Artificial Analysis measured Sonnet 5.5 in Claude Code: $3.33 per task at xhigh effort versus $14.19 at max.
Artificial Analysis
@artificialanlys
Sonnet 5.5 tops the Coding Agent Index at max effort, while xhigh offers an attractive price-performance tradeoff: 63 points versus 68 for less than a quarter of the cost. In Claude Code, Sonnet 5.5 from @AnthropicAI scores 68 at max versus 63 at xhigh. At $14.19 versus $3.33 per task, choosing xhigh cuts cost by 77% at the price of 5.5 index points. Max produces 4.8x more output tokens and takes 3.4x more turns. It improves all three benchmarks, most of all Terminal-Bench 4.0: 58.1% → 66.2%.
· 8.6K views
One agent, one task, two bills. A single effort setting makes the difference.
Artificial Analysis ran Sonnet 5.5 in Claude Code at two effort levels and published the results on Oct 14, 2026. At max, the model took first place in the Coding Agent Index with 68 points; at xhigh, it scored 63 and cost less than a quarter as much.
Where the savings come from. At max, the agent produces 4.8 times more output tokens and takes 3.4 times more turns. A task costs $14.19, while the same work at xhigh costs $3.33.
What you lose. Maximum effort lifts all three of the index's measurements, most of all Terminal-Bench 4.0: from 58.1% to 66.2%. The overall gap comes to 5.5 index points.
There is no free way into the agent: the entry Pro plan starts at $17 a month, and Max runs $100–200 (data as of Sep 26, 2026).
Cost per task is becoming as much a characteristic of an agent as its place in the index.