All insights

Video generation

Sora vs Veo: What Production API Teams Need to Know

Sora’s API was discontinued on September 24, 2026. See what that means for the Sora vs Veo decision, which Veo 3.1 output constraints are documented, and what teams should verify before choosing an API route.

Illustration for Sora vs veo

Direct answer

For a new production API integration, Veo is the viable choice of these two: OpenAI discontinued the Sora API on September 24, 2026. Google documentation lists a Veo 3.1 model on its Gemini Enterprise Agent Platform. Separately, Google says the Veo 3.1 preview models in the Gemini API were shut down on October 22, 2026. The Agent Platform and Gemini API preview are different access routes; don’t rely on old Gemini API preview instructions for a new integration. (OpenAI Help Center; Google AI for Developers; Google Cloud documentation)

Sora remains relevant as historical context—for example, when assessing existing Sora-generated assets or reviewing older comparisons. But historical prices, access routes, and test results do not establish a live Sora API option. Earlier articles also compared different pairings, including Sora 2 Standard versus Veo 3.1 and Sora 2 Pro versus Veo 3, so their conclusions should not be combined into a single current ranking.

Sora vs Veo for production APIs: comparison at a glance

Decision point Sora 2 Veo 3.1 on the Gemini Enterprise Agent Platform
Direct API status Discontinued September 24, 2026. Google documentation lists the model ID veo-3.1-generate-001.
Gemini API preview status Not applicable. Veo 3.1 preview models in the Gemini API were shut down October 22, 2026.
Documented output options No current Sora API output specifications to compare. The supplied model documentation lists video lengths of 4, 6, or 8 seconds and output resolutions of 720p, 1080p, or 4K. It does not establish here which length-and-resolution combinations apply to every generation mode.
Access and consumption notes No current direct Sora API access. The model documentation lists U.S. availability in us-central1, fixed quota, and Provisioned Throughput support; it lists pay-as-you-go as unsupported. Confirm current account and project conditions before planning around these details.
Practical decision Don’t plan a new integration around the discontinued direct API. Evaluate the documented model and its specific Google Cloud access, quota, and billing conditions for your workload.

Keep access routes separate. A consumer app, a direct provider API, and access through a third-party platform are not interchangeable. Historical third-party Sora access claims do not change the current direct API status. (OpenAI Help Center; Google AI for Developers; Google Cloud documentation)

Which model should you choose?

Choose Veo for a new video-generation API project

Between Sora and Veo, Veo is the actionable option for a new integration because the Sora API has been discontinued. Before committing, confirm that the documented Google Cloud route, quota, and billing arrangement fit your project. Google’s model documentation identifies veo-3.1-generate-001 and lists U.S. availability in us-central1; availability and account conditions should be checked for the project you plan to use. (OpenAI Help Center; Google Cloud documentation)

The available documentation excerpt lists 4-, 6-, and 8-second outputs and resolutions of 720p, 1080p, and 4K. It does not establish which combinations apply to every input type or generation mode. If duration, resolution, audio, reference inputs, or specific generation controls are compatibility requirements, confirm the exact combination in the current documentation before designing around it.

Treat Sora as a legacy comparison, not a live API contender

OpenAI’s original Sora 2 announcement described synchronized dialogue and sound effects. That historical product information may help interpret older reviews or existing assets, but it does not establish current API availability. The API discontinuation is the deciding factor for new direct API integrations. (OpenAI’s Sora 2 announcement; OpenAI Help Center)

For a legacy Sora workflow, inventory the inputs, controls, audio needs, output requirements, and downstream processing you depend on. Then test a current Veo route against representative tasks and acceptance criteria. Don’t assume the same prompt will produce equivalent results in Veo.

Production API considerations

  • Use the current route and model ID. Google’s Gemini API preview IDs are shut down. The supplied Agent Platform documentation lists veo-3.1-generate-001; don’t build a new integration around an older preview example. (Google AI for Developers; Google Cloud documentation)
  • Check output compatibility. The supplied documentation lists 4-, 6-, or 8-second video lengths and 720p, 1080p, or 4K resolutions. The available excerpt does not establish every supported length-resolution combination or how those options vary by generation mode. Confirm the exact configuration your use case requires. (Google Cloud documentation)
  • Confirm quota and billing arrangements. The supplied model-page excerpt lists fixed quota and Provisioned Throughput as supported, and pay-as-you-go as unsupported. Check the terms and quota applicable to your account rather than applying a historical per-second price from another route. (Google Cloud documentation)
  • Plan for lifecycle changes. The Sora API shutdown illustrates why model lifecycle matters when choosing a production dependency. Check the current lifecycle information for the exact model and access route before adoption; don’t rely on an unconfirmed retirement date.

This comparison does not establish service guarantees or predict generation reliability. Review the operational terms and documentation that apply to the specific route and account you intend to use.

Cost and access: compare like with like

There is no current direct Sora API price to compare because the API has been discontinued. Historical third-party articles reported prices for earlier model versions and different access routes; those figures are not current, like-for-like quotes for Veo on the Agent Platform.

For current Veo planning, identify the model ID, Google service and quota arrangement, and billing terms that apply to your project. The supplied Agent Platform documentation excerpt lists fixed quota and Provisioned Throughput for the standard model, but does not establish a directly comparable dollar rate. Don’t substitute a consumer subscription, an old Gemini API preview price, or a third-party platform price for the applicable provider terms. (Google Cloud documentation)

What earlier Sora-versus-Veo comparisons can—and cannot—tell you

Older comparisons can suggest useful test scenarios, but they do not provide a current production API benchmark. One supplied review compared Sora 2 Standard with Veo 3.1 and reported different outcomes across prompts. Another compared Sora 2 Pro with Veo 3, offering qualitative claims and suggested prompts rather than measured results. These are different model pairings and evidence types, so they should not be merged into one ranking.

For a Veo evaluation, run a small, task-specific pilot:

  1. Select representative tasks, such as a product turn, a speaking character, or a moving subject.
  2. Record the model ID, access route, prompt, inputs, output settings, and number of generations.
  3. Score results against your actual acceptance criteria, such as subject motion, continuity, audio intelligibility, and editing effort.
  4. Record failed generations and total cost under your actual billing arrangement.

This is a proposed evaluation method, not a test conducted for this article. Since Sora’s API is no longer available, it cannot serve as a new live control group. Use existing Sora outputs only if you already have comparable assets and records.

Bottom line

For a new production API decision, don’t choose Sora over Veo: OpenAI discontinued the Sora API on September 24, 2026. Veo 3.1 is the actionable option among these two in the supplied Google Cloud model documentation. Confirm the exact model route, output configuration, quota, and billing terms for your project. The Veo 3.1 Gemini API preview was also shut down on October 22, 2026, so distinguish the current route from older tutorials and comparisons. (OpenAI Help Center; Google AI for Developers; Google Cloud documentation)

Sources

What to know about the Sora discontinuation — OpenAI Help Center

Generate videos with Veo 3.1 in Gemini API — Google AI for Developers

Veo 3.1 — Gemini Enterprise Agent Platform documentation

Sora 2 is here — OpenAI

Contact us