AI Summary
The headline: Runway Gen-4.5 is #1 on the Artificial Analysis text-to-video benchmark as of November 2025, ahead of Google Veo 3 and OpenAI Sora 2.
Who built it: Runway, a 155-person New York studio founded by three NYU Tisch graduates in 2018. They co-released Stable Diffusion in 2022.
Who it’s for: Filmmakers, editors, ad agencies, and serious creators who want studio-grade output and the deepest creative tooling on the market.
Skip if: You want the cheapest-per-second video model (try Seedance), or you need a chat-style consumer entry point (try Gemini with Veo).
What is Runway Gen-4.5?
Google’s Gemini Omni Flash, announced May 19 2026, raises the bar on conversational video editing. Runway’s editor moat (the workflow, not the model) remains the strongest defense; the iteration-speed race is now the live story. See: Gemini Omni: Google’s Video Leap.
Gen-4.5 is Runway’s flagship text-to-video model, released December 1, 2025. It generates 5 to 10-second clips at high resolution, with native audio and what the company calls “believable collisions” — meaning when two objects interact in a scene, they behave with realistic weight and momentum instead of clipping through each other.
The benchmark that matters: 1,247 Elo on the Artificial Analysis Text-to-Video leaderboard as of November 30, 2025 — the highest score of any model. Runway moved past Google Veo 3 and the pre-shutdown version of OpenAI Sora 2 on that scoreboard, and they have held the top spot since.
Gen-4.5 is a closed-weights foundation model that runs through Runway’s web app, mobile app, and API. It is included on every paid plan at no premium — the same credit cost as the previous Gen-4 model.
Who built Runway, and how did they get here?
Runway was founded in 2018 by three classmates from NYU Tisch ITP: Cristóbal Valenzuela (CEO, from Chile), Alejandro Matamala-Ortiz (Chief Design Officer, also Chile), and Anastasis Germanidis (CTO, from Greece). They started in New York, near Union Square — not Silicon Valley — and they have stayed there.
The company’s under-credited fact: in August 2022, Runway co-released Stable Diffusion alongside the CompVis Group at LMU Munich, with backing from Stability AI. That release reshaped the open-source image-AI ecosystem. Most people associate Stable Diffusion with Stability AI; Runway was one of the three co-authors.
Funding has tracked the model progression closely:
| Round | Date | Amount | Valuation |
|---|---|---|---|
| Series C (lead) | Dec 2022 | $50M | ~$500M |
| Series C extension | Jun 2023 | $141M | $1.5B |
| Series D | Apr 2025 | $308M | ~$3.3B |
| Series E | Feb 2026 | $315M | $5.3B |
General Atlantic led both the 2025 and 2026 mega-rounds. The Series E investor list includes NVIDIA and AMD Ventures participating in the same round — rare for any AI company, and a signal Runway is positioning to be hardware-agnostic. Revenue tracked alongside: $80M ARR end of 2024, $300M ARR end of 2025, with another $40M added in Q2 2026 per Valenzuela.
What makes Gen-4.5 different from the previous generation?
Three capabilities matter most.
Physics-accurate motion. Gen-4.5 handles weight, momentum, and force well enough that objects collide, fluids splash, hair moves, and fabric drapes the way a film editor expects. This is the area where prior AI video models broke the illusion most often — people noticed when a glass tipped and the liquid moved wrong. Runway closed most of that gap.
Native audio. Like Veo 3.1, Gen-4.5 produces audio with the clip in a single generation — ambient sound, character voice (when prompted), music beds. The audio is generally less polished than Veo’s, but it eliminates the silent-clip problem and saves a round-trip through a separate audio model.
The A2D training technique. Gen-4.5 was trained on NVIDIA Hopper and Blackwell GPUs end-to-end using a technique Runway co-developed with NVIDIA called Autoregressive-to-Diffusion (A2D). NVIDIA CEO Jensen Huang publicly endorsed Gen-4.5 at launch, calling it “an incredibly exciting time for video and world models.” This co-development is part of why Gen-4.5 leapt the benchmark — and part of why NVIDIA stayed in as a Series E investor.
How much does Runway cost?
Runway uses a credit-based model on five tiers:
| Plan | Monthly | Annual (per mo) | Credits | Notes |
|---|---|---|---|---|
| Free | $0 | $0 | 125 one-time | No flagship video access |
| Standard | $15 | $12 | 625/mo | Gen-4.5 access |
| Pro | $35 | $28 | 2,250/mo | Higher res, longer clips |
| Unlimited | $95 | $76 | 2,250 + Explore Mode | Unlimited at relaxed rate |
| Enterprise | Custom | Custom | Custom | SSO, custom credits |
Annual billing saves about 20%. The most important pricing fact: Gen-4.5 access is included on every paid tier — Runway did not add a premium surcharge when the new model launched. If you were on Standard for Gen-4, you stayed on Standard for Gen-4.5 at the same price.
For most creators, the practical choice is between Standard ($12-15/mo) and Pro ($28-35/mo). Standard works if you generate 30-50 clips a month; Pro is where you go once you start iterating heavily or producing client work.
How do you actually use Runway Gen-4.5?
Three entry points, in order of how most people start:
The web app at runwayml.com. Sign up, log in, type a prompt, generate. Runway’s web app is the most polished creator interface in the AI video market — timeline editing, frame-by-frame control, in-browser preview, easy clip chaining. If you are coming from CapCut, DaVinci Resolve, or Adobe Premiere, this is the closest analog.
The mobile app. iOS and Android. Use it for casual prompting and quick captures-to-clips on the go. Editing is limited compared to the web app, but the rendering quality is identical.
The API. For teams building Runway into a product. Documented at runwayml.com/api. This is how enterprise customers integrate Gen-4.5 into ad-creative pipelines, marketing automation systems, and game prototyping tools.
Beyond the base video model, Runway has built four adjacent tools that matter for production work:
- Act-One / Act-Two — performance capture. Record yourself on a webcam, and Runway drives a generated character with your expressions and timing.
- Frames — an image model focused on stylized cinematography. Useful for storyboarding and concept frames before committing to video generation.
- Aleph — in-context video editor. Apply edits to existing video via prompt (“change the season to winter,” “replace the car”).
- Game Worlds — their first interactive world model, released December 2025. Simulates an environment and responds to user inputs, rather than just rendering a passive clip.
Who is actually using Runway in production?
Runway is one of the few AI video tools with a real Hollywood paper trail.
Everything Everywhere All At Once (2022, A24). VFX editor Evan Halleck used Runway’s rotoscoping and green-screen tools on the “rock universe” sequence. Halleck told Variety the tool was “cutting things out better than my human eye was, and it gave me a clean mat that I could use for other things.” Days of work compressed to minutes. To be clear about the scope: Runway was used in post-production, not for the generative VFX in the film.
The Late Show with Stephen Colbert. Uses Runway in TV editing workflows.
Lionsgate. Announced a September 2024 partnership to build a custom Runway model trained on Lionsgate’s catalog (John Wick, Hunger Games, Twilight). A one-year review from The Wrap in September 2025 reported that the catalog was still “too limited to train a good AI model.” The deal exists; public output does not.
AMC Networks. June 2025 — first cable network to formally partner with Runway, for marketing imagery and pre-visualization.
IMAX. Screened Runway AI Film Festival finalists in 10 commercial IMAX locations in August 2025.
Outside film, fintech is the surprise category. Chime, Robinhood, SoFi, and PayPal all use Runway for marketing video. The Runway pitch to brands: faster turnaround on creative-iteration cycles, lower per-asset cost than a traditional ad agency.
How does Runway compare to Veo, Kling, and Seedance?
| Model | Max clip | Native audio | Mid-tier API cost/sec | Distinctive |
|---|---|---|---|---|
| Runway Gen-4.5 | 10 sec | Yes | ~$0.50 (est.) | Best creator tooling, Hollywood adoption |
| Veo 3.1 | 8 sec | Yes | $0.40-$0.75 | Strongest audio, Google ecosystem |
| Kling 3.0 | 15 sec | Yes (multilingual) | $0.084-$0.168 | Character + voice carryover |
| Seedance 2.0 | 15 sec | No | $0.14 | Longest clips, cheapest per second |
Runway’s case rests on three things Veo cannot match today: deeper creator tooling (Frames, Aleph, Act-One), Hollywood production credits, and a focused company built around video first instead of bolted to a search engine. Veo’s case rests on audio quality, Google distribution, and cheaper entry tiers. For a full side-by-side see our Sora vs Runway vs Kling comparison (being updated to add Veo and Seedance) and the AI Video Generation Guide for 2026.
What are Runway Gen-4.5’s limitations?
Runway is honest about three categories of failure mode, all listed in their research blog:
Causal-reasoning errors. Effects sometimes appear before their causes — smoke before fire, splash before impact. This breaks the illusion in any clip where the order of events matters.
Object permanence issues. Things can disappear from a scene and not come back, or come back changed. A character’s necklace can vanish mid-clip. This shows up most in longer scenes where Runway is extending generation.
Success bias. When the prompt implies an action (“a basketball player shoots the ball”), the model is biased toward the ball going in. Realistic outcomes — misses, awkward landings, partial failures — require explicit prompting.
Beyond the model itself, two practical limits: 10 seconds per generation means anything longer is a chained sequence, and closed weights means you cannot run Gen-4.5 on your own infrastructure. If those are blockers, Runway is the wrong choice.
Get Smarter About AI Every Morning
Free daily newsletter — one story, one tool, one tip. Plain English, no jargon.
Free forever. Unsubscribe anytime.
Frequently Asked Questions
Is Runway Gen-4.5 free?
No, but the free tier gives you 125 one-time credits to test the platform. The cheapest plan with Gen-4.5 access is Standard at $12/month on annual billing (or $15/month). For most creators, Standard or Pro is the right starting point.
Can I use Runway videos commercially?
Yes, on all paid plans. Runway’s terms permit commercial use of generated content on Standard, Pro, Unlimited, and Enterprise tiers.
How long can Runway clips be?
Single generations run 5 to 10 seconds. Longer sequences are built by chaining generations — either through Runway’s extension tool or by treating each clip as a shot in a sequence and editing them together.
Does Runway have a watermark?
Generated clips on paid tiers do not include a visible watermark. Free-tier output sometimes includes one for content-disclosure purposes.
What is the difference between Gen-4 and Gen-4.5?
Gen-4 (March 2025) introduced consistent characters across shots and the Lionsgate custom model. Gen-4.5 (December 2025) added the audio capability, the physics improvements, and pushed Runway to #1 on the Artificial Analysis benchmark.
What are world models, and why does Runway keep talking about them?
A world model is a generative model that understands physics, causality, and object permanence well enough to simulate an interactive environment. Valenzuela has called AI video the “prequel” to world models — their thesis is that the same underlying technique powers robotics policy training, autonomous-vehicle simulation, and industrial design, not just creative clips. Runway released their first world model (Game Worlds) in December 2025 and announced a second one for robotics and medicine in their Series E announcement.
New market map: AI Short Drama Market Map 2026 — the full vertical-microdrama stack from foundation video models to creation tools to distribution platforms. Includes verified market data ($11B in 2025, $14B projected for 2026) and named tools across all three layers.
Sources
- Runway Research — Introducing Gen-4.5 (December 1, 2025)
- CNBC — Runway Gen-4.5 launch coverage
- TechCrunch — Runway company profile (May 15, 2026)
- Crunchbase News — Runway Series E ($315M at $5.3B, February 2026)
- Variety — Runway in Everything Everywhere All At Once post-production
- Runway — Pricing page
You might also like
- Google Veo 3.1 Explained — Runway’s closest competitor, with stronger audio.
- The Real State of AI Video in 2026 — our Special Report Vol. 2, with the full landscape.
- AI Video Generation Guide for Beginners (2026) — beginner walkthrough of every major AI video model.
- Every AI Model Worth Knowing in 2026 — 30+ models compared in one table.
- Sora vs Runway vs Kling — head-to-head of the leading AI video tools.
- AI for YouTube Creators — how creators are using Runway and other AI video models in production.
- AI Glossary — every term in plain English.