TokenMix Research Lab · 2026-05-18

Veo 4 in 2026: It's Not Released, So What Are You Buying?

Veo 4 in 2026: It's Not Released, So What Are You Buying?

Google DeepMind has not released Veo 4 as of May 12, 2026. Veo 3.1 is the publicly available latest. Yet platforms like veo4free.io are already selling "Veo 4" subscriptions at $29.90–$129.90/month. Here's what's real, what's wrapper, and what's actually coming.

Per DeepMind's official Veo page, the state-of-the-art model on display is Veo 3.1 — "Video, meet audio" — with no Veo 4 announcement, listing, or even a "coming soon" placeholder anywhere on the page. Independent industry coverage confirms this: Artlist's April 2026 analysis explicitly states "Google DeepMind has not officially announced Veo 4. All current information comes from public research trajectories, industry reporting, and the evolution of previous Veo models." Meanwhile, veo4free.io brands itself "Veo 4 — Free Multimodal AI Video Generator By Google DeepMind" — a claim that does not survive a 60-second audit. Its "About" page reads literally [Company Name] is dedicated to [brief description of what your company does]. Founded in [year]... — unfilled template placeholders. Its blog says "No blog posts." Its pricing tiers run $29.90–$129.90/month. The platform internally lists Seedance, Veo 3.1, Happyhorse, Nano Banana, and "Veo 4" as selectable models. Whatever "Veo 4" you would pay for there is not Google's, because Google's doesn't exist yet.

Table of Contents

Quick Answer

Question Direct answer
Is Veo 4 released? No. Google DeepMind has made no official announcement. Veo 3.1 is the current public model per deepmind.google/models/veo.
Then what is "Veo 4" on veo4free.io? A third-party platform using "Veo 4" as a product name. Their About page is unfilled template text. They internally route to multiple models (Seedance, Veo 3.1, Happyhorse, Nano Banana).
When will Google release Veo 4? No confirmed date. Industry consensus, per Artlist, points to sometime in 2026 based on Google's release cadence — but this is inference, not announcement.
What can I actually use today? Veo 3.1, Veo 3.1 Fast, Veo 3, Sora 2, Wan 2.6, Kling O1, Hailuo, Seedance — all real and shippable. Skip "Veo 4" subscriptions until Google announces.
Will Veo 4 beat Sora 2? Speculation. Predicted capabilities (4K, longer clips, stronger character consistency) suggest a competitive position, but no benchmarks exist for a model that has not been released.

Confirmed Facts as of May 12, 2026

# Fact Status Source
1 Google DeepMind's public Veo page lists Veo 3.1 as the latest model Confirmed deepmind.google/models/veo
2 Veo 3.1 description on the official page: "Video, meet audio. Our latest video generation model, designed to empower filmmakers and storytellers." Confirmed Same page
3 Google has not published any Veo 4 announcement, demo, paper, or release date Confirmed by absence DeepMind blog, Google AI blog, Cloud blog all searched
4 Artlist's analysis (last updated April 20, 2026) states Veo 4 "has not been officially released" Confirmed Artlist Veo 4 article
5 veo4free.io exists, presents "Veo 4 by Google DeepMind" branding, charges $29.90–$129.90/month Confirmed Direct site fetch May 12, 2026
6 veo4free.io About page contains unfilled template placeholders ("[Company Name]", "[year]", "[email protected]") Confirmed — significant trust signal veo4free.io/about
7 veo4free.io internally exposes multiple model selections including "Veo 4," "Veo 3.1," "Seedance," "Nano Banana" Confirmed Site UI inspection
8 Veo 4 will be released "in 2026" Likely (industry inference) Artlist analysis, Google's annual cadence
9 Veo 4 will support 4K, longer clip duration, stronger character consistency Speculation (industry expectation) Artlist's expectations grounded in Google research trajectory
10 "Veo 4" services available today are using Google's actual Veo 4 model No evidence. Reject claim by default. Google publishes no API access to a Veo 4 model

Veo 4 Status: What Google Has and Has Not Said

What Google has said about Veo 4: nothing. As of May 12, 2026, neither the DeepMind blog, the official deepmind.google/models/veo product page, the Google AI blog, the Google Cloud AI/ML blog, nor any Google research publication contains a Veo 4 announcement, capability document, or release window. The DeepMind Veo page's "Explore the latest" section promotes Veo 3 and Veo 3.1 exclusively. The Vertex AI product documentation for Google Cloud video generation references Veo 3 and Veo 3.1 model endpoints. A web search for site:google.com "Veo 4" returns zero indexed pages from Google domains.

What's actually been said about Veo 4 in 2026 publications:

The pattern this matches: Veo 3 was officially announced at Google I/O 2025 in May 2025. Veo 2 was announced in December 2024. Veo 3.1 arrived as a mid-cycle refresh in late 2025 / early 2026. If Veo 4 follows the same major-release cadence, Google I/O 2026 in late May is the most likely announcement window — meaning the model might be officially unveiled within days or weeks of this writeup. Until that announcement happens, any service selling "Veo 4" is selling something else under that label.

Veo 3.1: The Real Latest Model

Veo 3.1 is Google's actual state-of-the-art video model as of May 2026. Per the DeepMind product page, it ships through Gemini, Google Flow, and the Veo API. Its documented capabilities include:

Capability Veo 3.1
Max resolution 1080p
Max clip duration ~1 minute (pipeline-dependent)
Native audio Yes (ambience + simple dialogue)
Multi-shot continuity Yes
Character consistency across shots Limited — drifts on long sequences
Multilingual on-screen text Limited
Camera control prompts Yes (cinematic controls)
Image-to-video Yes
First-frame / last-frame conditioning Yes
Reference image support Yes

Veo 3.1 ships in three configurations: Veo 3.1 (full quality), Veo 3.1 Fast (lower latency, lower cost), and the older Veo 3 for legacy workflows. Pricing varies by access channel: Gemini app subscribers pay nothing direct; Google Flow uses credit packs; the Veo API on Vertex AI charges per-second of generated video, typically in the $0.30–$0.75 per second range depending on resolution and audio toggle. Most third-party aggregators (Artlist, Replicate, Pollo AI, Pika ecosystem) resell Veo 3 / 3.1 with a markup of 30–80% above Google's direct API cost.

Veo 3.1 is genuinely production-grade for short-form social content, storyboarding, previsualization, and rough cuts. Its honest weaknesses: 1080p ceiling, ~1-minute clip wall, continuity drift on multi-shot sequences, and audio that usually needs cleanup in post.

The "Veo 4" Wrapper Sites: A Field Investigation

A handful of domains have moved to claim the "Veo 4" name commercially before Google's release. The most prominent is veo4free.io. We fetched its homepage, About, pricing, and blog pages on May 12, 2026. The findings, presented without spin:

What veo4free.io claims

What our audit found

Our assessment

The most charitable read is that veo4free.io is a multi-model aggregator UI that's calling one of its internal model options "Veo 4" — possibly routing to Veo 3.1 in the background, possibly routing to a different vendor's model, possibly waiting to swap in real Veo 4 access if and when Google releases it. The less charitable read is that this is a brand-confusion product designed to capture searches for "Veo 4 free" and convert them into subscriptions before users notice that the real Veo 4 doesn't exist yet.

Either way: if you cannot verify which actual model is generating your video, you cannot evaluate its quality, cost-efficiency, or commercial usability. That is a problem you do not have on direct Google channels (Gemini, Vertex AI) or on transparent aggregators (Artlist, Replicate, Pollo AI) that disclose which model handles each request.

Predicted Veo 4 Capabilities (Industry Analysis)

These are predictions, not announcements. Based on Artlist's April 2026 analysis, published research from Google DeepMind on adjacent models (Nano Banana Pro, Gemini Robotics), and the upgrade pattern from Veo 2 → Veo 3 → Veo 3.1, the industry expects Veo 4 to deliver:

Predicted upgrade Industry confidence Why analysts expect it
4K resolution support High Most consistent creator demand; commercial work needs it; Veo 3.1 capped at 1080p
Longer clip duration (2-3+ minutes) Medium-High Veo 3.1 maxes around 1 minute with continuity drift; next leap addresses storytelling
Stronger character consistency High Google's Nano Banana Pro image model excels at this; expected to influence Veo 4 design
Multilingual on-screen text accuracy Medium-High Google's language model lead transfers naturally to video text rendering
Higher-fidelity audio with expressive speech Medium Veo 3.1's audio is timing-grade not delivery-grade; clear improvement vector
Reference sheet workflows (like Nano Banana Pro) Medium Powerful production workflow that would lift Veo into client-ready commercial use
Improved camera logic across cuts Medium Smooth shot-to-shot transitions remain a major Veo 3.1 weakness
Direct integration with Gemini agent workflows Speculative Possible given Google's agent push, but no signal yet

These predictions should not be treated as features you can rely on for production planning. Until Google publishes a model card, demo reel, and benchmark suite, these are educated industry guesses. A platform selling "Veo 4" today cannot deliver any of these capabilities — because the capabilities themselves don't exist in deployable form yet.

Veo Lineage: From Veo 1 to Veo 4 (Expected)

Model Release Key advance Status May 2026
Veo 1 May 2024 (Google I/O) First public DeepMind text-to-video; 1080p; up to 1 min Superseded
Veo 2 December 2024 Improved physics, camera control, longer prompts Available legacy
Veo 3 May 2025 (Google I/O) Native audio integration, multi-shot continuity, cinematic controls Available
Veo 3.1 Late 2025 / early 2026 Mid-cycle refresh; refined audio; better character consistency Latest available
Veo 3.1 Fast Concurrent with 3.1 Lower-cost, lower-latency variant of 3.1 Available
Veo 4 Not announced Predicted: 4K, longer clips, stronger continuity, expanded audio Not released

The cadence is clear: Google has shipped a major Veo release at I/O for two consecutive years. If that pattern holds, Google I/O 2026 (third week of May 2026) is the most plausible window for a Veo 4 official reveal. That window is days away from this writeup's publication date.

Veo 3.1 vs Sora 2 vs Wan 2.6 vs Kling O1: What You Can Use Today

While the market waits for Veo 4, four production-grade video models are shipping and routable today.

Dimension Veo 3.1 Sora 2 Wan 2.6 Kling O1
Vendor Google DeepMind OpenAI Alibaba Kuaishou
Max resolution 1080p 1080p Up to 4K (Wan 2.6 advances) 1080p
Native audio ✅ Yes Partial No (silent) No (silent)
Max clip length ~1 min ~20 sec ~10 sec ~10 sec
Multi-shot continuity Strong Strong Medium Medium
API access Vertex AI + aggregators OpenAI API (closed beta) Multiple aggregators Aggregators
Approx cost per second $0.30–$0.75 Pricing TBD post-shutdown $0.01–$0.05 (cheapest 1080p) $0.10–$0.25
Commercial use Yes (license terms apply) Yes Yes Yes
Best for Storytelling with audio Cinematic shots Cost-sensitive 1080p volume Stylized motion

The bottom-line trade-offs:

This is the actual decision space today, not "Veo 4 vs. everything else." On TokenMix.ai's model intelligence tracker we monitor pricing and availability across these video providers in real-time, so you can see when a Veo 4 endpoint actually appears in supported channels — and when it does, it will be on official Google domains and Vertex AI, not on a wrapper site with placeholder text.

5 Red Flags Before Subscribing to a "Veo 4" Service

Five concrete signals to check before paying any platform claiming Veo 4 access:

# Red flag Why it matters
1 About / Contact pages contain template placeholders Indicates the company never finished basic legal/disclosure setup. If they cut that corner, what other corners did they cut?
2 No published Google partnership or Vertex AI backend disclosure Real Veo access requires Vertex AI quotas or specific creator program partnerships. Real partners advertise the relationship; ghost partners don't.
3 Pricing in monthly "credits" with no per-second cost transparency Hides the underlying model identity. You can't compute cost-per-second or compare to direct API pricing if you don't know what model handled your request.
4 Multiple model options labeled with proprietary names If "Veo 4," "Seedance 2," and "Happyhorse" are all selectable, the platform is a routing layer. Routing layers are fine — but only if they disclose which model handles each request and what the underlying provider charges.
5 No public demo reel from your specific "Veo 4" tier A platform genuinely accessing Veo 4 would have demo content from Veo 4. Generic AI-generated samples with no model attribution are a substitute, not proof.

The simple rule: if you cannot identify the model that's actually serving your prompt, you are not buying Veo 4 — you are buying a wrapper that can swap its backend any time without telling you.

How to Try Veo Today: 3 Verified Paths

For users who want to test Google Veo capabilities right now, these are the three legitimate paths.

Path 1: Direct via Gemini app (consumer, easiest)

Sign in to gemini.google.com with a Google account. Gemini Pro and Gemini Ultra subscribers get Veo 3 / 3.1 access through the app's video creation feature. No API needed. Limits apply per subscription tier. Best for: prototyping, social content, single-creator workflows.

Path 2: Google Flow (creator-focused, more controls)

Google Flow is Google's dedicated creator tool exposing Veo with more direction controls — first/last frame, camera paths, multi-prompt sequencing. Uses credit packs. Best for: storyboarding, branded content, longer-form projects.

Path 3: Vertex AI Veo API (developer / production)

Vertex AI exposes the Veo 3 and Veo 3.1 models via API endpoints. Bearer token authentication, REST + SDK access, per-second pricing. Best for: programmatic generation, integration into agent workflows, production volume.

For pricing intelligence across Veo and competing video models, TokenMix.ai's research lab tracks per-second costs, latency benchmarks, and feature parity across Veo 3.1, Sora 2, Wan 2.6, Kling, Hailuo, Seedance, and emerging models so you can pick the right tool for each shot type. Through TokenMix.ai's unified API gateway, you can already route the 170+ language models — Claude, GPT-5.5, Gemini 3.1 Pro, DeepSeek V4 Pro, Qwen, Kimi, GLM, MiniMax — that handle the script writing, prompt engineering, captioning, and post-production tasks that surround any video generation workflow.

Final Recommendation

Do not subscribe to a "Veo 4" service in May 2026. Google has not released the model. Any platform charging you for Veo 4 access today is either (a) routing to a different, unspecified model, or (b) waiting to swap to real Veo 4 if and when Google releases it — without committing to refunds if the actual product differs.

If you want to use Google Veo today, the verified options are Gemini app, Google Flow, and Vertex AI — all at *.google.com or *.cloud.google.com. They run Veo 3.1, which is genuinely production-grade for storyboarding, social, and short-form work.

If you want the cheapest 1080p video generation today, Wan 2.6 is roughly 10–30× cheaper per second than Veo 3.1 for comparable resolution and clip length, at the cost of no native audio.

If you're waiting for Veo 4, watch DeepMind's Veo page and the Google AI blog directly. Google I/O 2026 in late May is the most plausible announcement window based on the previous two-year release cadence. When the real Veo 4 ships, the real source will be a google.com or deepmind.google URL — not a wrapper site with template placeholders on its About page.

FAQ

Has Google officially announced Veo 4? No. As of May 12, 2026, neither the DeepMind blog, the Google AI blog, the Google Cloud AI/ML blog, nor any official Google product page contains a Veo 4 announcement. The current latest is Veo 3.1.

Then what is veo4free.io selling? veo4free.io presents itself as a "Veo 4" platform with $29.90–$129.90/month pricing. Audit of the site shows its About page contains unfilled template placeholders, its blog has no posts, and its internal UI offers selection between multiple model labels including Seedance, Veo 3.1, Happyhorse, and Nano Banana — none of which are Google's actual Veo 4 (which is unreleased). We cannot verify what specific model handles requests under the "Veo 4" label there.

When will Veo 4 be released? No confirmed date. Google's historical I/O cadence (Veo 2 → Dec 2024, Veo 3 → May 2025) suggests Google I/O 2026 — late May 2026 — as the most plausible window. This is industry inference, not Google guidance.

Can I prepare my workflow for Veo 4 now? Yes, by using Veo 3.1 today through Gemini, Google Flow, or Vertex AI. Build character reference sheets, document prompt patterns, and design workflows assuming 4K timelines. Any skill you build on Veo 3.1 transfers when Veo 4 arrives.

Will Veo 4 beat Sora 2? No way to know. Predicted Veo 4 capabilities — 4K, longer clips, stronger audio — would be competitive, but no benchmarks exist for an unreleased model. Sora 2's API access situation also remains in flux, which complicates direct comparison.

Is Veo 3.1 worth using today? Yes for storytelling and social content where native audio adds value. The 1080p ceiling and ~1-minute clip limit constrain commercial use; for 4K-required work, wait for Veo 4 or use upscaling pipelines.

How do I verify a "Veo 4" service is legitimate? Check that the platform publishes a Google partnership disclosure, names the actual model serving each request, exposes per-second cost (not just monthly credit tiers), and points to demo reels with verifiable Veo 4 attribution. If any of these are missing, the platform is not selling Veo 4 — it's selling a wrapper.

What should I use if I need cheap 1080p video right now? Wan 2.6 leads on cost per second by a wide margin. See our Wan 2.6 review for the full pricing and capability comparison. Sora alternatives are also documented in our Sora API shutdown guide.

Related Articles

Sources


By TokenMix Research Lab · Published 2026-05-12 · Last Updated 2026-05-12 · Data Checked 2026-05-12