CompanyBench

Duplicate-pending complaint audit — CX hold-confusion triageCards & Interchange

13 verifiers · 7 LLM-graded

EnvironmentsJiraSlackZeta3 SQL

TaxonomyCards & InterchangeCards & Interchange

How each model performed on this task. Every chart states its N, and the reward carries a 95% confidence interval once there is more than one trial to measure.

Core metrics

Check Pass %

Mean share of the task's checks passing, averaged over trials. N = 10 trials per model.

  1. 68%

    GPT-5.5

  2. 67%

    Claude Opus 4.8

  3. 42%

    Gemini 3.1 Pro

  4. 33%

    Gemini 3.5 Flash

Extended metrics

Median Time

Wall-clock minutes, median across trials. N = 10 trials per model.

  1. 4m

    Gemini 3.5 Flash

  2. 3m

    GPT-5.5

  3. 3m

    Claude Opus 4.8

  4. 2m

    Gemini 3.1 Pro

Tool Calls

Tool invocations, median across trials — counted from the trajectories. N = 10 trials per model.

  1. 37

    Gemini 3.5 Flash

  2. 29

    GPT-5.5

  3. 18

    Claude Opus 4.8

  4. 18

    Gemini 3.1 Pro

Verifier × model pass rate

How often each verifier passed, per model. Green is passing, red is failing — a red band across a row is a verifier nothing clears.

VerifierGPT-5.5n=10Claude Opus 4.8n=10Gemini 3.1 Pron=10Gemini 3.5 Flashn=10
checks (13)
answer_complaints_reviewed100%100%20%100%
answer_harm_split_and_refund10%0%0%20%
deliverable_jira_ticket_filed100%100%50%20%
deliverable_slack_channel_exists100%100%90%20%
deliverable_slack_message_content0%0%0%20%
deliverable_slack_message_pinned100%100%90%20%
answer_incidents_and_holds60%40%80%40%
answer_reserved_total90%100%20%40%
floor_all_connectors100%100%50%20%
floor_sql_investigation100%100%100%100%
floor_write_back100%100%50%20%
trace_cohort_and_cluster_method100%100%100%60%
trace_hold_vs_cleared_reasoning100%100%100%40%

Failure modes on this task

38 tagged verifier failures, by the stage they were assigned to. Higher is worse.

Failure stageGPT-5.5Claude Opus 4.8Gemini 3.1 ProGemini 3.5 Flash
Answer1 mode1010108