CompanyBench

Confluence Overdraft & NSF PolicyRisk & Fraud

4 verifiers

EnvironmentsConfluenceJiraZeta3 SQL

TaxonomyRisk & FraudRisk & Fraud

Start with all trials for auto-written insights, or select a single trial to drill into its verifier results, trajectory, and workspace changes.

GPT-5.5

Mean reward

77.5%

All trials

Across 10 trials on this task, reward ran from 50% to 100% — 5 of 10 cleared every check.

  • 3.1 / 4avg checks passed
  • 32avg turns
  • 445avg tool calls
  • 19mavg time

Summary · GPT-5.5

GPT-5.5 passed 5 of 10 trials. Failures cluster on cp4_payday_nsf_transfers (5/10 trials).

Common pattern

cp4_payday_nsf_transfers failed in every unsuccessful trial; what else broke varies across the other 1 verifiers. One reproducible defect, with noise around it.

Why they broke

Most of GPT-5.5's failures on this task were classified as Wrong answer — 5 verifiers across 5 of 10 trials.

Per-trial notes

  • Trial 01Passed every verifier it was graded on.
  • Trial 02Missed 1 of 4 checks — cp4_payday_nsf_transfers.
  • Trial 03Missed 2 of 4 checks — cp3_healing_attempts and cp4_payday_nsf_transfers.
  • Trial 04Missed 2 of 4 checks — cp3_healing_attempts and cp4_payday_nsf_transfers.
  • Trial 05Passed every verifier it was graded on.
  • Trial 06Passed every verifier it was graded on.
  • Trial 07Missed 2 of 4 checks — cp3_healing_attempts and cp4_payday_nsf_transfers.
  • Trial 08Passed every verifier it was graded on.
  • Trial 09Passed every verifier it was graded on.
  • Trial 10Missed 2 of 4 checks — cp3_healing_attempts and cp4_payday_nsf_transfers.

Claude Opus 4.8

Mean reward

70%

All trials

Across 10 trials on this task, reward ran from 50% to 100% — 1 of 10 cleared every check.

  • 2.8 / 4avg checks passed
  • 29avg turns
  • 56avg tool calls
  • 4mavg time

Summary · Claude Opus 4.8

Claude Opus 4.8 passed 1 of 10 trials. Failures cluster on cp3_healing_attempts (7/10 trials).

Common pattern

No single verifier breaks every unsuccessful trial — failures move across 2 verifiers, so what varies between trials matters as much as the model.

Why they broke

Most of Claude Opus 4.8's failures on this task were classified as Wrong answer — 9 verifiers across 9 of 10 trials.

Per-trial notes

  • Trial 01Missed 1 of 4 checks — cp4_payday_nsf_transfers.
  • Trial 02Missed 2 of 4 checks — cp3_healing_attempts and cp4_payday_nsf_transfers.
  • Trial 03Missed 1 of 4 checks — cp3_healing_attempts.
  • Trial 04Missed 1 of 4 checks — cp3_healing_attempts.
  • Trial 05Missed 1 of 4 checks — cp4_payday_nsf_transfers.
  • Trial 06Passed every verifier it was graded on.
  • Trial 07Missed 1 of 4 checks — cp3_healing_attempts.
  • Trial 08Missed 2 of 4 checks — cp3_healing_attempts and cp4_payday_nsf_transfers.
  • Trial 09Missed 1 of 4 checks — cp3_healing_attempts.
  • Trial 10Missed 2 of 4 checks — cp3_healing_attempts and cp4_payday_nsf_transfers.

Gemini 3.1 Pro

Mean reward

92.5%

All trials

Across 10 trials on this task, reward ran from 75% to 100% — 7 of 10 cleared every check.

  • 3.7 / 4avg checks passed
  • 71avg turns
  • 102avg tool calls
  • 32mavg time

Summary · Gemini 3.1 Pro

Gemini 3.1 Pro passed 7 of 10 trials. Failures cluster on cp3_healing_attempts (2/10 trials).

Common pattern

No single verifier breaks every unsuccessful trial — failures move across 2 verifiers, so what varies between trials matters as much as the model.

Why they broke

Most of Gemini 3.1 Pro's failures on this task were classified as Wrong answer — 3 verifiers across 3 of 10 trials.

Per-trial notes

  • Trial 01Passed every verifier it was graded on.
  • Trial 02Passed every verifier it was graded on.
  • Trial 03Missed 1 of 4 checks — cp3_healing_attempts.
  • Trial 04Passed every verifier it was graded on.
  • Trial 05Missed 1 of 4 checks — cp3_healing_attempts.
  • Trial 06Passed every verifier it was graded on.
  • Trial 07Missed 1 of 4 checks — cp4_payday_nsf_transfers.
  • Trial 08Passed every verifier it was graded on.
  • Trial 09Passed every verifier it was graded on.
  • Trial 10Passed every verifier it was graded on.

Gemini 3.5 Flash

Mean reward

72.5%

All trials

Across 10 trials on this task, reward ran from 0% to 100% — 5 of 10 cleared every check.

  • 2.9 / 4avg checks passed
  • 57avg turns
  • 58avg tool calls
  • 17mavg time

Summary · Gemini 3.5 Flash

Gemini 3.5 Flash passed 5 of 10 trials. Failures cluster on cp4_payday_nsf_transfers (5/10 trials).

Common pattern

cp4_payday_nsf_transfers failed in every unsuccessful trial; what else broke varies across the other 3 verifiers. One reproducible defect, with noise around it.

Why they broke

Most of Gemini 3.5 Flash's failures on this task were classified as Missing field — 3 verifiers across 3 of 10 trials.

Per-trial notes

  • Trial 01Passed every verifier it was graded on.
  • Trial 02Passed every verifier it was graded on.
  • Trial 03Missed 1 of 4 checks — cp4_payday_nsf_transfers.
  • Trial 04Missed 4 of 4 checks — cp1_overdraft_offered, cp2_negative_balance_count and cp3_healing_attempts and 1 more.
  • Trial 05Passed every verifier it was graded on.
  • Trial 06Passed every verifier it was graded on.
  • Trial 07Passed every verifier it was graded on.
  • Trial 08Missed 4 of 4 checks — cp1_overdraft_offered, cp2_negative_balance_count and cp3_healing_attempts and 1 more.
  • Trial 09Missed 1 of 4 checks — cp4_payday_nsf_transfers.
  • Trial 10Missed 1 of 4 checks — cp4_payday_nsf_transfers.