Evaluations
Release, comparison and nightly runs for Support Assistant, scored by the rubric v3 judge. Select two to four runs to compare them side by side.
Runs, last 30 days
44
Nightly pass rate
87.2%
Best full golden-set run
86.9%
Release gate
Passed
107 of 107 runs
v3.4 · synthetic-hard-negatives v1.3 Experimentsynthetic-hard-negativesClaude Sonnet 5.5 340 / 540 RunningOct 8 | |
Nightly · v3.4 · 2026-10-08 Nightlysupport-golden-v4Claude Sonnet 5.5 87.2% | |
v3.4 · Claude Haiku 5.5 comparison Experimentsupport-golden-v4Claude Haiku 5.5 82.3% | |
Nightly · v3.4 · 2026-10-07 Nightlysupport-golden-v4Claude Sonnet 5.5 87.3% | |
v3.4 · Gemini 2.5 Pro comparison Experimentsupport-golden-v4Gemini 2.5 Pro 83.8% | |
support-assistant v3.4 · release candidate Releasesupport-golden-v4Claude Sonnet 5.5 86.9% | |
Nightly · v3.4 · 2026-10-06 Nightlysupport-golden-v4Claude Sonnet 5.5 86.6% | |
Nightly · v3.4 · 2026-10-05 Nightlysupport-golden-v4Claude Sonnet 5.5 86.8% | |
v3.4 · locale-formatting Experimentlocale-formattingClaude Sonnet 5.5 80.2% | |
Nightly · v3.4 · 2026-10-04 Nightlysupport-golden-v4Claude Sonnet 5.5 88.7% | |
Nightly · v3.4 · 2026-10-03 Nightlysupport-golden-v4Claude Sonnet 5.5 87.9% | |
v3.4 · billing-edge-cases v2.1 Experimentbilling-edge-casesClaude Sonnet 5.5 81.7% | |
Nightly · v3.4 · 2026-10-02 Nightlysupport-golden-v4Claude Sonnet 5.5 87.3% | |
Nightly · v3.4 · 2026-10-01 Nightlysupport-golden-v4Claude Sonnet 5.5 87.6% |