@Maceo shared a benchmark with me and
# generalv
Vaibhav (BAML 🐑)
07/15/2026, 2:20 PMUpdated benchmarks!
@Maceo shared a benchmark with me and i know @everyone has asked for an updated benchmark on BAML for a bit... turns our BAML still makes models quite good!
BAML makes gpt-oss-20b out perform gpt-5-mini.
(apologies in advance for the slop writing here, proper blog post incoming after I finish running on all models)
https://claude.ai/code/artifact/5c6626d0-fca9-474d-b9ef-92d3c9ae9c15
https://cdn.discordapp.com/attachments/1119375594984050779/1526956748454232175/Screenshot_2026-07-15_at_7.17.37_AM.png?ex=6a58e8ca&is=6a57974a&hm=0eb53582bbb5c0d7e242494052ea70e6c602c200aa14e7a526194d22ea75b912&
Vaibhav (BAML 🐑)
07/15/2026, 2:21 PMoriginal benchmark: https://interfaze.ai/blog/introducing-structured-output-benchmark
d
dex
07/17/2026, 5:00 PMthis is sick congrats
dex
07/17/2026, 5:00 PMnice work @Maceo
dex
07/17/2026, 5:01 PMyou should include the price! isn't 20b like 1/100 the cost of 5-mini?
dex
07/17/2026, 5:01 PMah 1/10th the cost not 1/100th
2 Views