Is Runway Down?

PRAXIS-1 -- RUNWAY'S FIRST OPEN-WEIGHT ROBOT POLICY MODEL, ANNOUNCED BUT NOT RELEASED (2026-09-30, vendor-primary): Runway Research announced **Praxis-1**, 'our first open-weight world action model', 'built on the same large-scale video pretraining behind our general world models' and positioned as 'a generalist policy model for robotics developers and researchers that works across any embodiment or environment'. **Status: 'We're actively testing Praxis-1 with early partners across a variety of embodiments, and will release it publicly in the coming months'** -- named partners are **Noble Machines, Standard Bots and Ultra**, each running it on their own hardware, with more partners getting pre-launch access. **No weights, licence, model size or pricing yet.** The research claims are vendor-reported: policy performance 'improves as we scale third-person video'; a chart puts final placement error after fine-tuning on web video at **16.1 cm vs 16.0 cm for teleoperated robot video** (error bars +/- 1 SEM over 93 evaluation pairs, 'differences smaller than a bar are not significant'), i.e. web video matches teleop data; simulating policies inside Runway's world model 'predicts real-world results with 0.95 correlation'. Failure cases Runway highlights as targets: repeated near-identical objects, clutter and occlusion, transparent objects, deformable cloth. **Why it is on this page:** it is the third Runway model in a month with no consumer surface (GWM Worlds 2 on 9/03, now Praxis-1), and the first Runway model promised as open weights. Nothing changes for Gen-4.5 users, plans or credits. Watch for the public release post and the licence; this is Runway's Robotics line, not video.

GWM WORLDS 2 -- RUNWAY'S REAL-TIME WORLD MODEL GAINS AUDIO AND SUBJECT/SCENE CONTROL, STILL A RESEARCH PREVIEW (2026-09-03, vendor-primary; recorded 2026-09-14 on a staleness check -- this page was last reviewed 8/06 and had only the May GWM-1 note): Runway published **GWM Worlds 2**, 'the next iteration of our research to create interactive environments.' **What it does:** 'continuous **720p video at 24 fps and audio at 48,000 Hz**, responding to your inputs as you explore'; built 'on top of our foundational audio-video generation model', it turns generation into real-time simulation where you 'steer the world with text actions addressed to any subject or to the scene itself, alongside continuous camera motion' -- a sword slash, a spoken reply, a dust storm -- with no preset session length. **The input format is the research contribution:** **WorldPrompt**, which splits world state into a persistent 'genesis' layer (scene description, subjects with attributes, 'laws' from gravity to character abilities, plus a first frame) and a time-varying action stream. Demos show first-person play, director-style scene prompts, and both at once. **Stated targets:** interactive entertainment, virtual characters, robotics/embodied-agent simulation, generative interfaces. **STATUS: research preview, published under runwayml.com/research, not a product feature and not in the Gen-4.5 app or API.** This extends the December GWM Worlds and the May GWM-1 family; it does not change pricing, plans or the Gen-4.5 leaderboard position. Watch for productisation at Runway's AI Summit (9/30, San Francisco), which the page promotes.

STALENESS FIX -- RUNWAY AGENT IS ON 2.0, AND IT IS NOW A MARKETING TOOL (2026-06-25, vendor; this page still described only the 5/13 Agent 1.0 launch): Runway shipped **Agent 2.0**, and the repositioning is the story. Agent 1.0 (5/13) was pitched as a general 'conversational AI creative partner' for planning and iterating video projects. **Agent 2.0 is explicitly aimed at marketers** -- vendor framing: 'Agent 2.0 helps marketers make more of what works: ads, videos and full campaigns that drive revenue.' Four named workflows: **performance marketers** upload existing creative plus ad metrics from **Meta, YouTube, TikTok or Google**, and Agent analyses what converted and generates the next set of ads to test; **social marketers** hand over a week of engagement data and get the next week of posts, auto-cut to **9:16 for Reels and Stories, 16:9 for YouTube, 1:1 for feed**; **brand marketers** develop campaign concepts conversationally and get localised copy and swapped visuals for new markets rather than rebuilding assets per region; **product marketers** work out positioning angles and then build campaign assets around the strongest messages. Read: Runway is moving Agent from 'creative tool' toward 'performance-marketing loop', which puts it against ad-creative platforms rather than only against other video models. Availability and pricing were not restated in the launch post -- Agent 1.0 was on all plans, and we have not found a first-party statement changing that, so do not assume 2.0 carries the same entitlement without checking

PLATFORM LAUNCH -- RUNWAY DEV + MEDIA ROUTER (2026-07-23, vendor; DATE CORRECTED 2026-08-03 -- this entry previously carried 2026-07-08, which was wrong: the vendor post is dated July 23, 2026). Alongside Dev, Runway shipped **Runway Media Router**, described by the vendor as 'the world's first model router for generative media models' -- it automatically picks the best video, image or audio model per request against user-defined preferences for cost, quality and latency. Runway launched **Runway Dev**, 'the AI media platform for developers: one API to integrate the best image, video, audio and real-time character models.' Four surfaces: **Models** (first-party Gen-4.5, Aleph 2.0, Act-Two PLUS third-party models incl. Seedance, GPT Image 2, ElevenLabs -- 'swap between them by changing one line of code'), **Recipes** (prebuilt endpoints: Ad Localization, Product Ad, Product Swap, Multi-Shot Video, Marketing Stock Image), **Workflows** (custom multi-model pipelines behind private API endpoints), and **Characters** (interactive avatars with voice, tool calling, knowledge bases). Enterprise posture: SOC 2 Type II, IP indemnification, no-training terms, 99.9% uptime, per-model cost visibility. Named customers: Adobe, ElevenLabs, Shutterstock, Figma Weave, Gamma. Pricing not in the launch post

MODEL SUNSETS -- ALL EXECUTED, THE API CALLS NOW FAIL (updated 2026-08-03): the sunsets are done, not upcoming. **Gen-3 Alpha went offline July 8, 2026** and **Gen-3 Alpha Turbo (`gen3a_turbo`) + Gen-4 Aleph (`gen4_aleph`) were removed on July 30, 2026** -- these are two separate dates, don't collapse them. Runway's API changelog, verbatim: 'July 30th, 2026 - Gen-3 Alpha Turbo (`gen3a_turbo`) and Gen-4 Aleph (`gen4_aleph`) are no longer available via the Runway API. Requests that use these model identifiers will fail.' The help center confirms: 'Gen-3 Alpha was retired on July 8, 2026, and Gen-3 Alpha Turbo was retired on July 30, 2026.' Both identifiers are now absent from Runway's live model list (verified 2026-08-03). Migration targets per vendor: **Gen-3 Alpha Turbo -> Gen-4.5** and **Gen-4 Aleph -> Aleph 2.0 (`aleph2`)** -- note there is **no 'Gen-5'** in Runway's docs, so ignore any source claiming one. If you still have pipelines pinned to Gen-3-era model ids, they are already broken

LATE-MAY/JUNE CLUSTER (all vendor-primary, runwayml.com/news): **5/26 Project Luxo** ('Crossing the Uncanny Valley of AI Media' -- research direction post), **5/27 Runway MCP** (drive Runway generation/editing from Claude, Cursor, and other MCP clients -- significant for agentic video pipelines), **6/1 Cosmos Coalition** launch, and **6/1 London HQ** opening as a world-model research hub for the UK/Europe. The MCP ship is the practical one: Runway becomes scriptable from inside coding agents, which no other major video platform offered as of early June