2.2.21

RyanCodrai/turbovec2.2.21Aug 6, 2026by github-actions[bot]

AI Summary

Includes a new SDK feature for bulk uploading experiment items and several backend performance optimizations.

Key Highlights

Add Experiment.batch_upload_items SDK feature for bulk operations
Optimize SELECT_BY_IDS performance by dropping FINAL spans
Register SambaNova as a canonical provider
Dedup spans by ID to improve data integrity

New Features

Experiment.batch_upload_items

Full Release Notes

## What's Changed
* [OPIK-7806] [QA] fix: harden model-picker and providers-table interactions in E2E POMs by @AndreiCautisanu in https://github.com/comet-ml/opik/pull/7770
* [OPIK-7534] Declare release-surface scope in the QA taxonomy by @AndreiCautisanu in https://github.com/comet-ml/opik/pull/7761
* [OPIK-7676] [BE] perf: drop spans FINAL from SELECT_BY_IDS (find_traces_by_ids) by @YarivHashaiComet in https://github.com/comet-ml/opik/pull/7719
* [OPIK-7469] [SDK] feat: add Experiment.batch_upload_items for the experiment items bulk endpoint by @JetoPistola in https://github.com/comet-ml/opik/pull/7726
* [NA] [BE] fix: register sambanova as a canonical provider so SambaNova model prices load by @Anuj7411 in https://github.com/comet-ml/opik/pull/7717
* [OPIK-7750] [BE] fix: dedup spans by id, dropping parent_span_id from the sort key by @thiagohora in https://github.com/comet-ml/opik/pull/7764


**Full Changelog**: https://github.com/comet-ml/opik/compare/2.2.20...2.2.21