Short answer: For a new integration through an official API, Veo 3.1 is the available option between these two: OpenAI says the Sora API was discontinued on September 24, 2026. Google documents Veo 3.1 models on Gemini Enterprise Agent Platform. That’s an API-availability answer—not evidence that Veo produces better video than Sora did. [1][2]
Status as of November 9, 2026: Google says its Veo 3.1 preview IDs in the Gemini API were shut down on October 22, 2026, and directs developers to GA models on Gemini Enterprise Agent Platform. Check the current model ID and product route; older Gemini API tutorials or third-party listings may describe a route that has been retired. [2][3]
Veo 3.1 vs Sora 2 at a glance
| Decision factor | Veo 3.1 | Sora 2 |
|---|---|---|
| Official API status | Google documents veo-3.1-generate-001 on Gemini Enterprise Agent Platform, as well as a separate Fast model, veo-3.1-fast-generate-001. |
OpenAI discontinued the Sora API on September 24, 2026. |
| Documented generation features | The standard-model page lists text-to-video, image-to-video, first-and-last-frame generation, video extension, reference images, and sound generation. | OpenAI’s launch announcement described synchronized dialogue and sound effects, improved physical realism, and controllability. Those are historical launch descriptions, not evidence of a currently available official API. |
| Standard-model output limits | 4-, 6-, or 8-second videos; reference-image-to-video supports only 8 seconds. Up to four output videos per prompt. Output resolution: 720p, 1080p, or 4K; 24 FPS. | No current official Sora API route for a like-for-like production comparison. |
| Documented access and consumption | Google lists U.S. availability in us-central1; fixed quota is supported and pay-as-you-go is not supported. Provisioned Throughput eligibility for this exact model is unconfirmed. |
Not a current option for a new official Sora API integration. |
| Comparable API price | The model specification does not state a per-second rate. | No current official Sora API price comparison is available after discontinuation. |
These are model-page specifications, not a quality benchmark. Google documents the standard and Fast models separately; for example, the standard model lists output up to 4K, while Fast lists up to 1080p. (docs.cloud.google.com) OpenAI’s launch announcement describes Sora 2’s original capabilities, while its discontinuation notice determines its current API status. [1][4]
Which model fits your workflow?
For a new official API integration: evaluate Veo 3.1
Veo 3.1 is the practical option between these two if you need a currently documented official API. Its standard model supports text-to-video and several image- or video-guided workflows, as well as sound generation. Those features may matter if your application needs to animate an image, specify start and end frames, or extend an existing Veo video. Note the mode-specific limit: reference-image-to-video on the standard model supports 8 seconds. (docs.cloud.google.com)
The documented consumption options include fixed quota, while pay-as-you-go is marked not supported for the standard model. The material checked for this comparison does not confirm whether veo-3.1-generate-001 is eligible for Provisioned Throughput, so don’t assume reserved capacity is available. Confirm your project’s quota, region, and commercial terms before choosing it for a workload. (docs.cloud.google.com)
For a prior Sora workflow: plan a migration
OpenAI’s original Sora 2 announcement described synchronized dialogue and sound effects, physical realism, and controllability. Those claims can help explain older comparisons, but they do not establish a current production advantage: OpenAI says the Sora API was discontinued on September 24, 2026. If another platform advertises Sora 2, check which provider operates that route and what service terms apply; it is not the discontinued official OpenAI API. [1][4]
For a visual-quality decision: don’t infer a current winner
Older third-party comparisons make qualitative claims about realism, cinematic style, continuity, and audio. They are not a substitute for a current controlled comparison, and some compare different versions, such as Veo 3 with Sora 2 Pro. Google’s model specification documents Veo features and limits; it does not prove that Veo will outperform Sora on a particular prompt. The discontinued official Sora API also prevents reproducing that exact comparison through OpenAI’s API today.
Instead, evaluate the active model against your own task—for example, animating a product image, creating a short dialogue scene, or extending a shot. Set acceptance criteria in advance and record prompt adherence, continuity, audio usefulness, usable results per attempt, and cost under the access route you can actually use. That is a task-specific Veo evaluation, not a new Veo-versus-Sora benchmark.
Production API access: what is documented—and what still needs confirmation
For Google Cloud, the documented standard-model ID is veo-3.1-generate-001. Google identifies Veo on Gemini Enterprise Agent Platform as an API route and documents text-to-video generation. The REST references describe long-running publisher-model predictions, and the video input schema defines a text prompt. [2][5][6]
A production integration needs to submit a generation request, retain its operation identifier, check for completion, and retrieve the resulting video. Google documents generated-video results as Cloud Storage URIs. [8][9]
Implementation details to confirm before writing or deploying a REST request: The material retrieved for this comparison does not establish a complete Veo-specific REST example—namely, the exact fully qualified request URL, the precise JSON envelope and required fields for this model, the authentication header format for that request, the applicable polling URL and API version, or whether and where to set an output-destination parameter. Do not treat an assembled URL or sample JSON as a verified, executable request. Confirm those details in the current Google documentation for the chosen model and route. Google’s general Agent Platform setup documentation discusses Application Default Credentials and API keys, but that alone does not establish the exact authentication header for a particular Veo request. [7]
This is distinct from using a consumer app, subscription, or third-party aggregator. Access through one of those routes does not establish access to Google’s official Agent Platform API.
What will Veo 3.1 cost?
There is no current apples-to-apples official API price comparison with Sora 2: OpenAI’s Sora API is discontinued. For Veo 3.1, Google’s model page lists fixed quota as a supported consumption option and pay-as-you-go as not supported for the standard model. It does not state a per-second or per-video rate, and the material checked here does not confirm this model’s eligibility for Provisioned Throughput. [1][2]
Google describes Provisioned Throughput generally as reserved capacity with commitment-based pricing. That general description does not establish that this specific Veo 3.1 model is eligible. The sources reviewed here also do not establish the price, minimum commitment, or exact billing basis that would apply to a particular Veo 3.1 project. A reliable estimate such as “one eight-second clip costs $X” therefore cannot be calculated from the model specification alone. Confirm the applicable arrangement with Google for your project, region, and throughput needs, then estimate using expected generations and retries. Keep API costs separate from consumer subscriptions and third-party platform prices. [2][11]
Bottom line
- Starting a new official integration? Evaluate Veo 3.1 on Gemini Enterprise Agent Platform, then confirm project access, quota, region, and price.
- Maintaining an official Sora API workflow? Plan a migration: OpenAI says the Sora API was discontinued on September 24, 2026. [1]
- Choosing based on output quality? The official documentation cited here does not establish a current head-to-head winner. Test the available route against your own acceptance criteria, and confirm its commercial terms before budgeting.
Sources
What to know about the Sora discontinuation — OpenAI Help Center
Veo 3.1 — Gemini Enterprise Agent Platform documentation
Generate videos from text prompts — Gemini Enterprise Agent Platform
Method: models.predictLongRunning — Gemini Enterprise Agent Platform
Get started with Gemini Enterprise Agent Platform
Long-running operations — Gemini Enterprise Agent Platform
VideoGenerationModelResult — Gemini Enterprise Agent Platform
VideoGenerationModelInstance — Gemini Enterprise Agent Platform