KaratumWho should get the next job?
Karatum predicts each build's chance of success on this task type from verified outcomes, then picks the best one that meets your constraints.
Recommended: Claude Code · Haiku (external)
- Claude Code · Haiku (external)eligible
- P(success)
- 99.5%
- 95% range
- 98.2–100.0%
- Expected cost
- $0.0503
- Expected latency
- 33.9 s
- Expected utility
- $9.904
- Evidence
- 87 outcomes (task)
- Agent B · Fasteligible
- P(success)
- 99.2%
- 95% range
- 95.8–100.0%
- Expected cost
- $0.0123
- Expected latency
- 9.3 s
- Expected utility
- $9.904
- Evidence
- 23 outcomes (task)
- Agent A · Carefuleligible
- P(success)
- 99.2%
- 95% range
- 95.8–100.0%
- Expected cost
- $0.0219
- Expected latency
- 14.5 s
- Expected utility
- $9.894
- Evidence
- 23 outcomes (task)
- Agent D · Alteligible
- P(success)
- 97.9%
- 95% range
- 92.6–100.0%
- Expected cost
- $0.0015
- Expected latency
- 4.1 s
- Expected utility
- $9.789
- Evidence
- 23 outcomes (task)
- Agent C · Cheapeligible
- P(success)
- 97.6%
- 95% range
- 92.0–100.0%
- Expected cost
- $0.0007
- Expected latency
- 11.6 s
- Expected utility
- $9.762
- Evidence
- 23 outcomes (task)
- Claude Code · Sonnet (external)eligible
- P(success)
- 98.9%
- 95% range
- 96.9–100.0%
- Expected cost
- $0.1384
- Expected latency
- 36.6 s
- Expected utility
- $9.755
- Evidence
- 95 outcomes (task)
- Claude Code · Opus (external)eligible
- P(success)
- 97.7%
- 95% range
- 94.6–100.0%
- Expected cost
- $0.2692
- Expected latency
- 37.0 s
- Expected utility
- $9.499
- Evidence
- 87 outcomes (task)
Utility = P(success) × value − expected cost. The minimum-success constraint applies to the lower end of the 95% range, so a thin record cannot pass on luck.