Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- Gemini API released gemini-omni-1.1-flash GA on August 27, 2026, recommending it as the replacement for gemini-omni-flash-preview.
- The changelog says the old endpoint will be deprecated on September 30. The deprecations table lists that date as the earliest possible shutdown date, not an observed shutdown. The exact date will be announced separately.
- Version 1.1 adds video extension, first/last-frame interpolation and resolution controls for 360p, 720p, 1080p and 4k. One text-to-video test is not a complete migration.
- Regress request schemas, asynchronous job states, output downloads, quality, latency, cost and failed retries.
- Retain a controlled rollback switch while the old endpoint works, and a stop plan for when rollback is impossible. Do not wait until the deadline approaches for the first bulk switch.
Interpreting the two official pages
Gemini API release notes mark gemini-omni-1.1-flash GA and say gemini-omni-flash-preview will be deprecated on September 30, 2026. The deprecations table labels that date a shutdown date, but its introductory note says all dates are the earliest possible shutdown dates, with exact dates announced separately in advance. The September 28 review observed neither an actual shutdown nor a specific shutdown time. Complete migration validation early without presenting the table date as an executed outcome.
Possible symptoms include the old model disappearing, model not found, asynchronous jobs failing to start or third-party platforms still hardcoding preview. Blind retries or network changes are the wrong first response. Confirm the actual model field and entry point in request logs.
Mapping the old and new models
The old model ID is gemini-omni-flash-preview and the new one is gemini-omni-1.1-flash. Search application code, environment variables, model routing, low-code connectors, Batch files, scheduled jobs and database configuration for the old string, recording each caller's owner and traffic.
Do not conflate a latest alias, the AI Studio selector and a pinned API ID. Acceptance records should show the model actually sent in requests and actual model information returned in responses or job objects.
New 1.1 GA features do not prove automatic compatibility
Google lists three additions: extend jobs or prompt-driven video extension, first/last-frame interpolation with up to two images, and video_config.resolution support for 360p, 720p, 1080p and 4k. The default is 720p; 1080p and 4k are upsampled.
These establish feature scope, not byte-identical compatibility with old requests and output. Systems depending on fixed dimensions, duration, first/last frames, job states or vendor wrappers require individual regression checks. One successful response cannot establish a complete migration.
Seven migration steps from inventory to staged release
In order: freeze new old-model configurations; export the caller inventory; fix representative inputs and expectations; switch to 1.1 in testing; validate generation, editing, polling, cancellation and downloads; release to a small traffic share; then update default routing and unattended jobs. Preserve request IDs, states, latency, usage, cost and artifacts at each step.
Also check whether SDKs, third-party gateways and internal model allowlists recognize the new ID. If the upstream supports it but an intermediary rejects it, do not misdiagnose the intermediary's schema error as a Gemini outage.
Accepting video quality, resolution and costs
Cover at least text generation, single-image animation, conversational editing, extend, first/last-frame interpolation and different resolutions. Compare composition stability, text, character continuity, duration, resolution, processing time, failure rate and cost per successful artifact.
4k and 1080p are upsampled outputs. Pixel dimensions alone do not establish native high-resolution detail. Use current model pages, project limits and bills for prices, quotas and regions; this page does not estimate undisclosed per-task costs.
Error layers and network boundaries
model not found, incompatible fields and unsupported tasks concern models or schemas. 429 concerns quota. For 5xx, preserve the request ID and follow official retry guidance. Investigate networks only with connection-timeout, DNS, TLS or proxy-authentication evidence. Changing exits cannot restore a retired model.
Use the Gemini 429/503 guide and AI API configuration guide to identify the layer before deciding to retry, fall back or escalate support.
Stop, rollback and revisit shutdown notices
Stop expanding traffic if critical quality regresses, cost or latency exceeds limits, asynchronous states are incompatible, output downloads are lost, or a vendor wrapper lacks the new ID. Short-term rollback is possible only while the old endpoint actually works. After shutdown, stop affected tasks and use another verified supported path; retries of the old ID cannot restore it.
Search old-model usage again, separating production traffic, background jobs and unowned configurations. Recheck when later official notices and your project's request results appear. If callers remain unknown, keep monitoring and old configuration records, limit impact and assign manual takeover.
Sources
Frequently Asked Questions
When will gemini-omni-flash-preview stop working?
Google's deprecations table lists September 30, 2026 as the earliest possible shutdown date; the exact date will be announced separately in advance. As of September 28 there was no evidence of an actual shutdown. Validate migration immediately.
What is the recommended replacement model ID?
Google recommends gemini-omni-1.1-flash, the GA version released on August 27, 2026.
Is replacing the model string enough?
No. Regress generation, editing, asynchronous polling, cancellation, downloads, resolution, quality, latency, cost and third-party wrappers.
Which resolutions does 1.1 support?
Google lists 360p, 720p, 1080p and 4k. 720p is the default; 1080p and 4k are upsampled.
Is model not found on the old model a network issue?
First check model ID, retirement status and intermediary allowlists. Investigate networks only with evidence such as DNS, TLS or connection timeouts.
Can a failed migration always roll back to Preview?
No. Controlled short-term rollback is possible only while the old endpoint remains available. After actual shutdown, stop old tasks and use another verified supported path.