feat(ai-gemini): add Gemini Omni 1.2 Flash (gemini-omni-1.2-flash) #2065
Send to my agentClosed
tombeckenham opened this issue ·0 comments
Google released Gemini Omni 1.2 Flash (gemini-omni-1.2-flash). Docs: https://aistudio.google.com/docs/omni
Omni 1.1 stays available for four weeks, then from 2026-12-03 Google routes it to 1.2 and bills at 1.2 rates.
What changes in 1.2#
- Duration range is 3 to 15 seconds (1.1 is 3 to 10). Default 10.
- Pricing: $12.00 per 1M video output tokens, about $0.07 per second at 720p (1.1 is $17.50 and $0.10). 360p $0.023/s, 1080p $0.104/s, 4k $0.209/s.
- New
generation_config.video_config.task: 'extend'appends 3 to 10 seconds per turn, up to 30 seconds total. - Reference limits: up to 6 reference images (3 per character, 2 characters). Up to 3 reference videos of 3 seconds for 3 to 10 second outputs, only 1 for 11 to 15 second outputs, none with
editorextend. First/last frame pair only for 3 to 10 seconds. - Audio input and voice editing are still unsupported.
- Control strings
<FIRST_FRAME>,<LAST_FRAME>,<IMAGE_REF_0>,<VIDEO_REF_0>are deprecated. The adapter does not emit them.
Work#
-
packages/ai-gemini/src/model-meta.ts: addGEMINI_OMNI_1_2_FLASHnext to 1.1 with the new pricing, add it toGEMINI_VIDEO_MODELSandGEMINI_INTERACTIONS_VIDEO_MODELS, mark 1.1 as deprecated in its doc comment. -
packages/ai-gemini/src/video/video-provider-options.ts: duration entries for 1.2 with max 15, mentionextendin the Omni provider options comment. -
scripts/sync-provider-models.ts: addgemini-omni-toNON_CHAT_MODEL_PREFIXESso the daily sync stops inserting Omni as a chat model. The sync PR #2063 already inserted a bogus chat entry for 1.2 that must be removed. - Docs:
docs/adapters/gemini.mdanddocs/media/video-generation.mdmodel tables, the Omni section indocs/media/video-generation-advanced.md(15 second limit,extend, reference count rules),docs/config.jsondates. - E2E fixture and spec for the new model id, changeset for
@tanstack/ai-gemini.
tanstack-source added labelsfeature
AlemTuzlak closed this
Sign in to join this conversation.