Sandbox
Comfortable
Prompts
prompt_suppo...
Tour
Prompts
support-reply
prompt_support_reply
v19
current best candidate for promotion
Compare latest
Linked traces
Experiments
Copy ID
Versions
3
Latest
v19
Traces
5
Error rate
40.0%
Avg cost
$0.0490
Version history
v19
current best candidate for promotion
4/22/2026 · gpt-4o-mini
v18
cheaper and faster, but less reliable on edge cases
4/22/2026 · gpt-4o-mini
v17
last known good before the rollout
4/22/2026 · gpt-4o
Version 19
gpt-4o-mini · Created 4/22/2026, 4:10:16 AM
Diff vs v18
current best candidate for promotion
Recovery candidate with restored policy grounding
Impact: v18 → v19
Regression
Error rate
100.0%
→
0.0%
-100.0pp
Avg cost
$0.0392
→
$0.0418
+7%
Avg latency
9.25s
→
9.65s
+4%
Traces
2
→
2
Same
Changes from v18
Full diff view
Compressed low-cost
Recovery
rollout
candidate
with
refund
restored
regressions
policy grounding