2.13.1 (2026-05-11)

Fix

• fix: pin jina omni vllm-free revisions (#4653)

fix: pin jina omni revisions without vllm dependency

Use HF revisions whose remote code passes optional dependency checks without requiring vllm.

Co-authored-by: Cursor <cursoragent@cursor.com> (9a5dac2)

Unknown

• model: jina-embeddings-v5-omni models (#4604)

• add: jina-embeddings-v5-omni models

Co-authored-by: Cursor <cursoragent@cursor.com>

• fix: route jina omni models through multimodal wrapper

Co-authored-by: Cursor <cursoragent@cursor.com>

• fix: load jina omni nano in fp32

Co-authored-by: Cursor <cursoragent@cursor.com>

• test: use mock task for jina omni prompt dispatch

Co-authored-by: Cursor <cursoragent@cursor.com>

• fix: pin jina omni revisions with whitespace parity

Co-authored-by: Cursor <cursoragent@cursor.com>

• fix: pin jina omni latest private revisions

Co-authored-by: Cursor <cursoragent@cursor.com>

• fix: simplify jina omni review follow-up

Move the nano dtype default into model metadata and remove model-specific tests per reviewer guidance.

Co-authored-by: Cursor <cursoragent@cursor.com>

Co-authored-by: Cursor <cursoragent@cursor.com> (5ff08fa)

• Activate statistics check for beta tasks (#4647)

update test (67a3032)

• data: add stats for V/VA pair classification tasks (#4646) (667cfa5)

• fix tolerate off-by-one frame metadata in compute_video_hashes (#4625)

fix: tolerate off-by-one frame metadata in compute_video_hashes (018d1ed)