Runway Gen-5

Runway 旗下 AI 视频生成与世界模型产品线，当前官方在售旗舰为 Gen-4.5，Gen-5 尚未由官方正式发布

video generation world model

https://runwayml.com/

Visit Website

In-depth Report

Runway is an AI video generation company headquartered in New York. It was founded in 2018 by three alumni of the ITP program at NYU Tisch School of the Arts. It is currently one of the leading players in the Vincentian video track. A key fact needs to be stated in advance: As of July 2026, the video model officially promoted by Runway is Gen-4.5, which was released in December 2025. Neither the official website nor the research report page has any official model named "Gen-5" online. The name "Runway Gen-5" currently mainly appears on third-party aggregation/alliance landing pages and unconfirmed industry rumors, and is not an officially confirmed product by Runway. This report is based on Runway's real products and company facts, and has a separate section to clarify the existential disputes of "Gen-5" to avoid taking rumors as established facts.
Runway (full name Runway ML) was co-founded by Cristóbal Valenzuela (CEO), Anastasis Germanidis (CTO) and Alejandro Matamala-Ortiz (CDO) and is headquartered in New York. The company debuted with the Gen series video model in the early days, and iterated from Gen-1 and Gen-2 in 2023 to Gen-4 and Gen-4.5, establishing the product category of "generating videos with text or pictures". At the capital level, Runway is one of the companies with the most intensive financing in the AI video track in 2025–2026. In April 2025, the company completed a US$308 million Series D financing, with a valuation of approximately US$3.3 billion, led by General Atlantic, with participation from SoftBank, Nvidia, Fidelity, etc. On February 10, 2026, Runway completed another US$315 million in Series E financing, with its valuation rising to approximately US$5.3 billion. This round was continued to be led by General Atlantic, and heavyweight institutions such as Nvidia, Adobe Ventures, AMD Ventures, and Fidelity followed in the investment, with a total financing of more than US$860 million. According to Crunchbase News, Runway serves tens of millions of consumer and business users and has annual recurring revenue in the hundreds of millions of dollars. The company's positioning has also been upgraded from a "video generation tool provider" to a "world model research company." In 2026, Runway launched the GWM-1 universal world model, which includes three variants: GWM Worlds (explorable environment), GWM Avatars (conversational characters), and GWM Robotics (robot operations), and launched the Runway Characters real-time video agent API based on GWM-1. In addition, Runway has established cooperation with NVIDIA (Vera Rubin Architecture Collaboration), Lionsgate Pictures (film preview and storyboarding), UCLA School of Film and Television, etc.
Since "Gen-5" is not yet an officially confirmed product, this section is based on Runway's current real flagship Gen-4.5, which is also the ability that users can actually use. Gen-4.5 was released on December 1, 2025, internally codenamed "David" (also known as Whisper Thunder), and is built on NVIDIA Hopper and Blackwell GPUs. Officials claim that it has reached the current highest level in terms of motion quality, prompt word compliance and visual fidelity, and ranks first on the Artificial Analysis Vincent video list with an Elo score of 1,247, surpassing competing products such as Google Veo 3 and OpenAI Sora. Gen-4.5 can generate about 5-10 seconds of video at a time, and supports one-click upgrade to 4K resolution export, which can directly enter Premiere, Final Cut, DaVinci and other editing processes. In terms of specific capabilities, Gen-4.5 emphasizes three points: first, physical realism, objects have credible weight, momentum, and collision reactions, and liquids and materials are dynamic and natural; second, complex command understanding, a single prompt word can specify camera movement, scene composition, event time, and atmosphere changes; third, a wide range of styles, from photorealism, film quality to stylized animation, all of which maintain a unified visual language. In terms of control methods, Gen-4.5 supports various modes such as Vincent video, Picture-based video, key frame control, and Video-based video. In addition to the Gen series, Runway's current lineup also includes models such as the Gen-4, Gen-4 Turbo, Aleph 2.0, Act-Two (action performance), and a host of AI magic editing tools (remove background, erase and replace, motion tracking, smart frame filling, etc.).
This is the key point that must be clarified in this report. As of July 2026, Runway’s official website (runwayml.com) and research page still mainly recommend Gen-4.5 and GWM-1. There is no official release record of “Gen-5” in the official changelog and announcements. The so-called "Gen-5" mainly appears in three types of places: The first is third-party aggregation/alliance landing pages, such as runwaygen5.org, aitools.fyi, topai.tools, etc., which package multiple third-party models such as Veo 3, Sora 2, Seedance, Wan 2.5, Kling, etc. for sale under the "Runway Gen-5" brand. These are not official Runway products. The second is rumors and reports from industry media. For example, there is an article saying that "Gen-5 Full is priced at US$0.15/second, Gen-5 Lite is US$0.08/second, and supports up to 8K." This type of statement lacks an official source of Runway, and the same media (promptyze) itself has issued a warning. In March 2026, it was circulated that "Gen-5 will be released on March 3, with a 90-second upper limit and delays due to timing consistency issues. Statements such as "July" cannot be verified in any authoritative source, which is a typical AI information distortion. Third, some evaluation/SEO content directly applies Runway’s general capabilities (character consistency, reference frames, key frame control) to the “Gen-5” name. Therefore, if users or readers understand Gen-5 as "the official next-generation model on sale", it is not currently valid; a safer approach is to regard Runway Gen-4.5 as a real product that can be evaluated at the moment. Once Runway officially releases Gen-5, the official announcement and changelog should prevail.
Runway adopts a subscription system based on "credits". Points are consumed based on model, duration, and resolution. Gen-4.5 consumes 12 points per second of video (60 points for 5 seconds and 120 points for 10 seconds). The official current packages are as follows: Free version: 125 points are given as a one-time gift (not expired), but does not include Gen-4.5 video generation, and the output is watermarked. The Standard version has a monthly payment of US$15 (annual payment is equivalent to US$12/month), 625 points per month, and is available for all models, watermark removal, and 4K enlargement. The Pro version costs $35 per month ($28 per month annually), costs 2,250 points per month, and adds custom voices (lip sync/TTS). The unlimited version Unlimited/Max costs US$95 per month (annual payment is US$76 per month), 9,500 points per month, and comes with unlimited generation of Explore Mode (relaxed queue, single line may wait 10–20 minutes during peak periods). The enterprise version is a customized quotation that provides compliance capabilities such as SSO, data not used for training, and workspace analysis. For serious creators, the 625 points of the standard version are approximately equivalent to 25 seconds of Gen-4.5 video, which are often used up in a few sessions; real daily production requires at least Pro or Unlimited. API is also subject to speed limits based on points prepayment and usage grading. The overall pricing is at a mid-to-high level among mainstream Wensheng video tools, and the gap is widened by model breadth and enterprise features such as 4K export and watermark removal.
After the release of Gen-4.5, the community generally recognized its improvements in action quality and prompt word restoration. Creators have reported that the complex lens scheduling, realism of physical movement, and consistency of long hair and materials in motion are significantly better than the previous generation; the production efficiency of film and television previews, advertising storyboards, and social media short films has been significantly improved. Third-party evaluations (such as pxz.ai Chinese evaluation) claim that the scene creation speed can be increased by about 90% and the preliminary preparation time can be shortened by about 70%. The negative feedback focuses on three points: first, the free version does not open Gen-4.5 and has watermarks, and the experience gap is obvious; second, the points are not carried forward (the monthly points of the standard version/professional version are reset every month), which is not friendly to users with low frequency but occasional large demand; third, although the Explore Mode is advertised as unlimited, the queue wait is too long during peak periods, and the real-time performance is limited. Additionally, as with all Vincent video tools, support for complex multi-agent, long-duration, lip-sync/dialogue audio is still limited.
Runway has long been regarded as a technical benchmark in the Vincent video circuit. Gen-4.5 topped the Artificial Analysis list, surpassing Google Veo 3, and was interpreted by Chinese media such as the Securities Times as a landmark event in the "battle for the video model throne." At the end of 2025, competing products such as Kering O1, Google Veo 4, and OpenAI Sora 2 have intensively followed up, and the industry has entered a stage of multi-power competition. However, Runway still maintains a strong voice due to its model breadth, corporate cooperation (Lions Gate, NVIDIA) and "world model" research narrative. Crunchbase regards Runway’s Series E as the epitome of “the popularity of capital in the video AI track”—financing for global AI video companies will increase by about 95% year-on-year in 2025.
In addition to the aforementioned “Gen-5” naming confusion, Runway faces several types of risks. The first is the risk of information distortion: Alliance pages and rumors in the name of "Gen-5" may mislead users to pay for non-existent models or form wrong expectations. It is recommended that all purchases and evaluations be based on the official website. The second is copyright and compliance risks: The source of training data, portraits and copyright boundaries of AI-generated videos are still industry-wide disputes, and corporate users need to be cautious when producing branded content. The third is the risk of the competitive landscape: Keling, Veo, Sora, Seedance, Pika, etc. are iterating very fast, and Runway’s top position is not ironclad. The fourth is cost risk: the points consumption and waiting time for heavy use may exceed the expectations of small and medium-sized teams.
If you are an independent creator, marketing team or film and television pre-show staff, Runway Gen-4.5 is one of the most worth-trying Vincent video tools currently. It is recommended to start with the Pro version to balance points and cost. This is especially suitable for brand content and strong character consistency needs (mascots, series). If you only have occasional needs or have a tight budget, you can first use the free version to verify the effect, or compare with lower threshold solutions such as Keling and Pika. Regarding "Gen-5", it is recommended to be cautious: do not pay directly for any third-party aggregation page labeled "Runway Gen-5", treat it as "the next generation that has not been officially announced"; when you really want to evaluate Runway's current capabilities, please use Gen-4.5 and GWM-1 world models as objects. If Runway officially releases Gen-5 in the future, priority should be given to checking the official changelog and benchmark data.
Runway is an AI video and world model company with solid technology and strong capital. Its current real-life flagship Gen-4.5 is in the first echelon of quality and rankings. However, as of July 2026, "Runway Gen-5" is still a name mainly based on rumors and third-party aggregation pages, and has not been officially confirmed. There is a risk of misjudgment in treating it as a next-generation product on sale. For users, the most pragmatic judgment is: evaluate now with Gen-4.5 and wait for Gen-5 with official announcements.

User Reviews

km5ukcn

—

Explore Mode 说是无限生成，但高峰期一条等十分钟以上，真赶工的时候还是挺焦虑的。

Peter.CastilloJr

—

从投资人视角看 Runway 这轮 E 融资挺有意思。53 亿美元估值、英伟达 Adobe AMD 全进场，说明资本认的不只是它现在的视频模型，更是「世界模型」这个更大的故事。GWM-1 把视频生成往实时仿真、机器人操作、对话角色三个方向铺开，一旦跑通，Runway 就从工具公司变成底层平台。当然风险也明摆着：可灵、Veo、Sora 迭代极快，榜首不是铁王座，而且 AI 视频的版权和训练数据争议始终悬在头上。短期看 Runway 技术声量还在，长期能不能撑住估值，得看世界模型能不能真正落地变现。

SpencerCook

—

4K 放大导出进 Premiere 很顺，工作流衔接比很多竞品干净，这点对剪辑师是真加分。

枫叶_14

—

入门先免费版试效果，认真做生产直接上 Pro 或 Unlimited，别在 Standard 上纠结，积分根本不够造。

b90pj9

—

复杂多主体和长时长还是短板，口型对话也别指望，真要精细表演还是得后期手动补。

安然_14

—

看到网上一堆「Runway Gen-5」的聚合页，点进去发现是把 Veo、Sora 一堆模型打包卖，根本不是官方的东西，大家别被名字骗了。

PEwal88

—

企业版才承诺数据不用于训练，这点对品牌客户很关键，普通套餐还是得掂量下数据边界。

Olivia.Reed_2023

—

E 轮估值 53 亿美元、英伟达和 Adobe 都进来了，资本这么站台，文生视频这条赛道短期热度下不来。

Terry.Howard369675

—

用 Runway Gen-4.5 做了一支产品预告片的分镜预演，镜头运动和物理真实感确实比上一代强太多，头发和布料在运动里都不糊了。

Judy.CastilloQ

—

积分每月清零这个设计对低频但偶尔爆量需求的用户太不友好了，希望官方能像 Max 那样给个结转。

Cynthia_RodriguezII

—

Artificial Analysis 榜单上 Gen-4.5 登顶，超过 Veo 3，这个成绩对做交付的团队来说就是说服力。

90utyhe399

—

作为一个从 Gen-2 就开始玩 Runway 的老用户，我想认真聊聊它和竞品的真实差距。Gen-4.5 在文生视频榜单登顶确实不是水军刷的，物理真实感和提示词遵循度肉眼可见地好，尤其物体重量感和液体动态，比同期的可灵、Pika 更自然。但别神话它：长视频、多主体、对口型这三块依然是行业通病，Runway 也没解决。另外价格真心不便宜，Standard 那点积分就是体验券，真干活至少 Pro 起步。我的建议是把它当生产力工具而不是玩具，想清楚自己要产出什么再决定买哪档。

Lauren.Patel_Pro

—

我们团队拿 Runway 和狮门的合作案例当参考，给客户做电影预演，效率和成本都漂亮，客户很买账。

Stephen_Edwards_66

—

把 Runway 当「现在的 Gen-4.5」来评就对了，至于 Gen-5 真要来了，等官方 changelog 说话再信不迟。

海浪_23

—

图生视频加关键帧控制很好用，IP 角色跨镜头一致性稳，做吉祥物和连续短剧省了好多重绘的功夫。

Kyle.Ward_9962

—

关于网上铺天盖地的「Runway Gen-5」，我专门去官网和 changelog 翻了一遍，结论是截至目前官方根本没有发布任何叫 Gen-5 的模型，主推的还是 Gen-4.5 和 GWM-1 世界模型。那些标着 Gen-5 的落地页基本是第三方把 Veo、Sora、Seedance 等一堆模型打包重卖，甚至同一篇媒体自己都发过文章打假。所以遇到「Gen-5 0.15 美元一秒」这类说法，别急着信，等 Runway 自己的公告。真要评估 Runway 当下能力，认准 Gen-4.5 就行。

dury1

—

GWM-1 世界模型和 Runway Characters 实时视频智能体这套叙事挺超前，感觉 Runway 想做的已经不止视频工具了。

毛素_1

—

我们工作室主要做 IP 吉祥物的短视频矩阵，最看重的是角色跨镜头一致性，这方面 Runway Gen-4.5 确实给了惊喜。以前用别家，换个景别脸就歪了，要后期大量修；现在图生视频加参考帧，同一只吉祥物在十几个镜头里发型、衣服、配色都稳得住，省下的重绘工时非常可观。不过也有坑：积分消耗比想象快，一条十秒 4K 就上百积分，批量做吉祥物日更内容得算好预算。总体对我们这种强角色需求的团队，Runway 是目前最稳的选择，前提是钱包扛得住。

JackJenkinsJr

—

我是在一家二十人左右的广告公司负责内容生产的，上个月把 Runway 接进了日常分镜和社媒物料流程。最直观的感受是 Gen-4.5 出片稳定性比早期模型高了一大截，以前一条复杂运镜要反复抽卡五六次，现在两三次就能定。但成本控制是硬伤，Pro 的 2250 积分在提案季根本撑不住，我们后来直接上了 Unlimited，单月 ninety-five 美元其实比按条外采渲染便宜。算总账的话，Runway 帮我们把预演环节的人力砍掉近一半，客户改需求也能当天出新版，这块省下的钱远超订阅费。

SandraRiveraK

—

Gen-4.5 的提示词还原度是真的好，一条长提示就能把运镜、构图、氛围都安排明白，省了我大半沟通成本。

JMurphy_2021

—

Pro 版每月 2250 积分，做日常社媒短片够用，但一旦要批量出片很快就烧光，重度用户还是得上升 Unlimited。

Susan.Moore_X

—

免费版连 Gen-4.5 都不给，还带水印，体验断层太明显了，新用户第一次打开很容易劝退。

Runway Gen-5

In-depth Report

User Reviews

Similar Products