ObviousBench

PUBLIC MODEL RELIABILITY · CONFIGURATION DETAIL

Claude Opus 5

Observed settings from the 2026-07-25 public release. This is a configuration readout, not a claim about general model quality or a vendor guarantee.

Open in the cost frontier Open share image
Lowest-cost observed 100% result100%

High · $0.5998

Settings observed6

ordered deepest to lightest

Benchmark sample144

items · 3 answers each

Observed answer pass3 by reasoning settingSettings are ordered from deeper to lighter / no reported reasoning.
Claude Opus 5 observed pass³ and full-run cost by setting 100% 95% 90% 70% 99.3%Max$0.7376 99.3%Extra high$0.6583 100%High$0.5998 99.3%Medium$0.503 98.6%Low$0.433 80.6%None / disabled$0.2455

Point labels show answer pass3 and estimated full-run cost. The table below retains the exact released observations.

RELEASED OBSERVATIONS

All visible settings

Answer pass3 means all three sampled answers were correct. Cost is the estimated public API cost for a full benchmark run.

SettingAnswer pass3Full-run costReasoning evidence
Max99.3%$0.7376Reasoning reported
Extra high99.3%$0.6583Reasoning reported
High100%$0.5998Reasoning reported
Medium99.3%$0.503Reasoning reported
Low98.6%$0.433Reasoning reported
None / disabled80.6%$0.2455No reported reasoning

How to read this page

These values are a release-bound observation set, not an estimate of every task or production workflow. Compare the full public field and global Pareto reference before making a cost or reliability decision.

Compare every public configuration on the cost frontier →