WorkloadBenchFollow @BrenBuilds on X

Claude Sonnet 5.5

Claude Sonnet 5.5's own checks, newest first. Its results on the tests every model ran are on the Quick suite and Deep suite pages.

Claude Code 2.1.318, 4 November 2026#

Sonnet 5.5 on Claude Code 2.1.318

Claude Sonnet 5.5 had the same typical scores on Claude Code 2.1.318 as on 2.1.284. Every output count overlapped.

Source: WorkloadBench at workloadbench.brenhq.com, updated 4 November 2026. 6 runs on Claude Code 2.1.318 on 4 November 2026, and 6 on 2.1.284.

Copy link
Notes

Typical means the middle run, or the midpoint of the two middle runs.

Both releases ran at medium effort on the same tasks and settings. The 2.1.284 runs are Claude Sonnet 5.5's from 27 October 2026.

They ran on other days, so this check can't tell the release apart from the day. Three runs per model and task on 2.1.318, so a small difference can be chance.

Output tokens are Claude Code's own count for the model. Hidden tests are tests the model never sees.

Joined deep suite v3, 26 October 2026

Claude Sonnet 5.5 joined version 3 on 26 October 2026, after the first two models' runs of 15 and 16 October 2026. It averaged 1.00 on every correctness task, and 2.56 on speed-up, 2.53 on scheduling and 0.52 on forecasting on log2 scales. 4 of its 12 runs hit the time limit.

Every model's runs are on deep suite v3.

Launch day on the explain task, 28 September 2026

These ran before 30 September, with each CLI's defaults left on, so their numbers differ from the Quick suite's explain task table.

Sonnet 5.5 against Sonnet 5 and Opus 5.5#

Claude Code's first request held 40,414 tokens for Sonnet 5 and 28,127 for Sonnet 5.5, and Sonnet 5 took three steps where Sonnet 5.5 took two.

Sonnet 5.5, Sonnet 5 and Opus 5.5 on the explain taskEach bar is the median of 2 runs, on Claude Code 2.1.284 at its default effort, medium for Sonnet 5.5 and Opus 5.5, high for Sonnet 5.

Total tokens#

Everything the model read and wrote in the whole run

Sonnet 5.5: 42,884 tokens, median of 2 runs, mostly light runsSonnet 5.542.9k*Sonnet 5: 115,842 tokens, median of 2 runsSonnet 5116kOpus 5.5: 36,465 tokens, median of 2 runs, mostly light runsOpus 5.536.5k*

Cost at list prices#

What the run's tokens cost at the model's published list prices, cached tokens at the cache price

Sonnet 5.5: $0.0602 at list prices, median of 2 runs, mostly light runsSonnet 5.5$0.0602*Sonnet 5: $0.110 at list prices, median of 2 runsSonnet 5$0.110Opus 5.5: $0.109 at list prices, median of 2 runs, mostly light runsOpus 5.5$0.109*

Tokens read#

All the text sent to the model, counted each time it was sent

Sonnet 5.5: 41,973 tokens, median of 2 runs, mostly light runsSonnet 5.542.0k*Sonnet 5: 115,001 tokens, median of 2 runsSonnet 5115kOpus 5.5: 34,984 tokens, median of 2 runs, mostly light runsOpus 5.535.0k*

Read from cache#

Share of what it read that came from the cache, which is cheaper

Sonnet 5.5: 75% of prompt tokens, median of 2 runs, mostly light runsSonnet 5.575%*Sonnet 5: 82% of prompt tokens, median of 2 runsSonnet 582%Opus 5.5: 74% of prompt tokens, median of 2 runs, mostly light runsOpus 5.574%*

Cache writes#

Text saved to the cache for reuse

Sonnet 5.5: 11,233 tokens, median of 2 runs, mostly light runsSonnet 5.511.2k*Sonnet 5: 20,610 tokens, median of 2 runsSonnet 520.6kOpus 5.5: 9,248 tokens, median of 2 runs, mostly light runsOpus 5.59.2k*

Output tokens#

Everything it wrote: the answer, its tool calls and its hidden thinking

Sonnet 5.5: 911 tokens, median of 2 runs, mostly light runsSonnet 5.5911*Sonnet 5: 842 tokens, median of 2 runsSonnet 5842Opus 5.5: 1,482 tokens, median of 2 runs, mostly light runsOpus 5.51.5k*

Hidden thinking#

The part of the output spent thinking, which the answer doesn't show

Sonnet 5.5: 37 tokens, median of 2 runs, mostly light runsSonnet 5.537*Sonnet 5: 27 tokens, median of 2 runsSonnet 527Opus 5.5: 424 tokens, median of 2 runs, mostly light runsOpus 5.5424*

* At least half the runs behind a starred number are light. Light runs start with fewer tokens of instructions and tools than the model's other runs, so their counts differ. What light runs are.

The table scrolls sideways on a small screen.

Same Claude Code release, explain task, no skill, medians
ModelPassedTokens readFrom cacheCache writesOutputThinkingVisibleCost at list prices
Claude Sonnet 5.5Claude Code 2.1.284, medium effort, Claude Code's default2 of 241,973*75%*11,233*911*37*875*$0.0602*
Claude Sonnet 5Claude Code 2.1.284, high effort, Claude Code's default2 of 2115,00182%20,61084227815$0.110
Claude Opus 5.5Claude Code 2.1.284, medium effort, Claude Code's default2 of 234,984*74%*9,248*1,482*424*1,058*$0.109*