Fireworks AI · feature_addedOn-policy RL for GLM 5.2 via Training API

valid_from

2026-06-24

valid_to

null (still holds)

confidence

0.70

superseded_by

null

first seen

2026-10-28

last updated

2026-10-28

Sources

“Run on-policy RL on GLM 5.2 with batch-invariant serving and zero-KLD train/serve alignment through the Fireworks Training API. See the validation results.”— what the source said

• Frontier RL Training Infrastructure for GLM 5.2primary · fireworks.ai

/claim/clm_B8C9F94BF2D0.md/v1/claims/clm_B8C9F94BF2D0timelinesubmit a correction