For a new production API workflow, evaluate Veo 3.1—not because current head-to-head tests prove it makes better video, but because OpenAI says the Sora API was discontinued on September 24, 2026. Google’s Gemini Enterprise Agent Platform documentation lists Veo 3.1 as generally available. This status summary is current as of November 8, 2026. (OpenAI’s Sora discontinuation notice; Google’s Veo 3.1 documentation)
The versions and access channels matter: older reviews compare Sora 2 with Veo 3, Sora 2 with Veo 3.1, or Sora 2 Pro with Veo 3.1. Those names and findings aren’t interchangeable, and tests conducted through a third-party platform aren’t production API benchmarks.
At a glance
| Decision point | Sora | Veo |
|---|---|---|
| Production API status | OpenAI says the Sora API was discontinued September 24, 2026. (OpenAI) | Google lists the Veo 3.1 model veo-3.1-generate-001 as generally available on Gemini Enterprise Agent Platform. (Google Cloud) |
| Other access routes | OpenAI says the Sora web and app experiences ended April 26, 2026. (OpenAI) | The separate Veo 3.1 preview in the Gemini API shut down October 22, 2026. That preview is distinct from the Agent Platform model documented as GA. (Google AI Studio; Google Cloud) |
| U.S. model region documented | No current Sora API route to evaluate, according to OpenAI’s notice. | Google lists U.S. availability in us-central1 for veo-3.1-generate-001. (Google Cloud) |
| Consumption options and price | Not applicable to a discontinued Sora API. | The model page lists fixed quota and Provisioned Throughput as supported; pay-as-you-go and batch inference are not supported for this model. The page does not provide a comparable per-video price. (Google Cloud) |
Bottom line: If you need an API now, evaluate Veo 3.1 through the documented Google Cloud route. If you’re asking which model produced better creative results in historical tests, the available evidence is too limited to name a universal winner.
What Veo 3.1’s production documentation says
For model veo-3.1-generate-001, Google documents text-to-video, image-to-video, generation from first and last frames, video extension, reference-image workflows, sound generation, and Content Credentials (C2PA). The model supports 4-, 6-, or 8-second videos; image-to-video with reference images is limited to 8 seconds. Documented aspect ratios are 9:16 and 16:9, input resolutions are 720p and 1080p, and output resolutions are 720p, 1080p, or 4K. The documentation lists sound generation as supported, while audio input is not supported. (Google Cloud)
These are documented specifications for Veo 3.1 on Agent Platform—not proof that Veo performs better than Sora did. They also shouldn’t be assumed to apply to another Veo version, a different access route, or a third-party platform.
For U.S. deployments, the model page lists us-central1 and a quota of 50 regional online prediction requests per minute per base model. It lists fixed quota and Provisioned Throughput as supported consumption options, and pay-as-you-go and batch inference as unsupported for this model. Check your project’s applicable quota and consumption arrangement before committing to an implementation. (Google Cloud)
What older Sora 2 vs Veo 3 tests showed
A third-party review published October 8, 2025 compared Sora 2 and Veo 3 through Artlist, not through their production APIs. Its author says the clips were generated at 720p and scored the models on three scenes:
| Test scene | Sora 2 | Veo 3 | Reviewer’s result |
|---|---|---|---|
| Breakdancing nun: action and prompt coherence | 5/5 | 2/5 | Sora 2 scored higher in this generation. |
| Origami birds: visual appeal and composition | 2/5 | 4/5 | Veo 3 scored higher in this generation. |
| Cyborg handshake: motion | 2/5 | 3/5 | Veo 3 scored higher in this generation. |
| Total | 9/15 | 9/15 | A tie in this small, subjective test. |
Treat that as a record of one reviewer’s results—not a repeatable benchmark or a forecast of API performance. The review discloses Artlist-sponsored test credits and affiliate links. It also notes an exception to its stated same-prompt method: for the handshake scene, the author removed the audio portion after both models failed to make the Milky Way move. (Review and methodology)
Another third-party article compared Veo 3.1 with Sora 2 Pro and recommended different models for different workflows. But the captured article says it tested identical prompts without reporting outputs or scores, and gives conflicting estimates for Veo 3.1 Fast’s speed advantage. It cannot establish a winner either. (Wiro comparison)
Which should you choose?
- Building a new production API integration: Evaluate Veo 3.1 through the documented Gemini Enterprise Agent Platform route. OpenAI says the Sora API has been discontinued, so there is no current Sora API to compare against it. (OpenAI; Google Cloud)
- Choosing on creative quality alone: Don’t infer a universal winner from the older tests. They cover a few subjective 720p generations through a third-party platform, not a controlled API benchmark. Compare the exact versions and routes available to you using matched prompts, settings, and repeated generations.
- Comparing costs: Keep API usage separate from app subscriptions and third-party credits. The old Artlist review’s “competitive pricing” description and annual-plan promotion don’t establish a current API price; neither do qualitative compute-cost claims in the older articles. The available sources don’t support a like-for-like per-generation price comparison.
Verdict
For production API selection as of November 8, 2026, Veo 3.1 is the option to evaluate among these two because OpenAI says it has discontinued Sora’s API. For which model makes better videos, the October 2025 Sora 2 vs Veo 3 review ended in a tie, but that result is not evidence of current model performance. Confirm the exact model version, API route, region, quota, and consumption arrangement before committing to an implementation.
Sources
What to know about the Sora discontinuation — OpenAI Help Center
Veo 3.1 — Gemini Enterprise Agent Platform documentation
Generate videos with Veo 3.1 in Gemini API — Google AI Studio