Today we are officially launching Coactive Multimodal AI Platform, Autumn '25 edition.‍

This isn’t just another update. It’s a fundamental re-imagining of how companies can use multimodal AI to create business value from video and image content. For media companies and other content-driven organizations, video is both a tremendous asset and a complex challenge. MAP Autumn '25 helps to turn that challenge into opportunity.

Video: an Opportunity Hiding in Plain Sight

Video is both the rising and dominant form of content online. It drives the majority of global internet traffic and has become the preferred way for people to consume information or entertainment.

But while video libraries are growing exponentially, much of the potential value of video remains inaccessible. The problem isn’t a lack of content—it's a lack of understanding of that content.

If you don’t know what’s inside your video library, you can’t monetize it effectively, personalize it for viewers, or ensure that it’s brand-appropriate. Manual tagging and traditional keyword search methods simply can’t keep up. Foundation models alone, while powerful, don’t address the complex operational needs of enterprises working at scale.

What’s New in MAP Autumn '25

We designed the latest version of MAP with customer outcomes in mind: faster discovery, more accurate tagging, lower costs, and smarter workflows, built for enterprise needs. Here’s how:

1. Faster, More Accurate Content Discovery (Video and Image Search)

Autumn '25 introduces tiered video search results. Instead of returning a large volume of results, the platform prioritizes the most relevant matches first, while also surfacing deeper recommendations with one click.

Combined with hybrid search signals—blending visuals and metadata—users can now find specific people, logos, scenes, moods, or dialogue with high accuracy. You can quickly locate relevant images or shots of a particular athlete smiling across hours and hours of content. 

2. Automated Video and Image Metadata Tagging at Scale

One of the most powerful upgrades in MAP Autumn '25 is the improvements to tagging accuracy. We’ve boosted zero-shot tagging accuracy to 81%, nearly double earlier benchmarks. That translates directly into fewer manual corrections, higher confidence in metadata quality, and faster time-to-value.

We’ve also added high-leverage tagging workflows. Users can tag a small subset of assets and quickly scale across larger datasets. Built-in LLM-powered prompts help refine tags, while auto-generated ground truth reports give teams visibility into quality. The result: metadata that’s richer, more reliable, and more deterministic.

3. Streamlined Content Workflows Across Teams

MAP Autumn '25 isn’t just about better results; it’s also about smarter processes. Tagging is now possible at every level: keyframe, shot, scene, and full video. Teams can preview results, share reusable tag templates, and eliminate redundant work.

For large organizations, this means cross-department alignment. Editorial teams, ad ops, and compliance reviewers can all work from the same metadata layer, saving time and ensuring consistency.

Coactive can help drive revenue, while reducing costs and removing process friction 

What does all of this mean for customers? The results speak for themselves:

• 50% higher tagging accuracy: Customers are seeing dramatic improvements in how quickly and reliably they can structure their libraries, so that they can capture the most value.
  
  
• 30–60% lower ingestion costs: With new efficiencies, companies can process up to 2,000 video hours every hour dependent on specific conditions—while spending less.
  
  
• Direct revenue impact: One media customer recently generated an additional $500,000 in ad revenue from a single advertiser by using Coactive to validate the quality and context of ad placement.
  
  

New capabilities that change the game

Beyond core improvements in speed and accuracy, MAP Autumn '25 introduces several new features designed to empower customers:

• Upgraded Tag Refinement: Fine-tune metadata with ease using LLM-powered prompts, intuitive previews, and auto-generated reports.
  
  
• VIP Identification: Easily search for or tag the very important people, products, or premises (concepts) that are relevant to your organization, with advanced named-entity recognition. Easily identify emerging celebrities, branded products, or key concepts, such as “female empowerment.”
  
  
• Content Safety Controls: Built-in safeguards to automatically screen for unsafe, biased, or toxic content—protecting brand suitability at scale.
  
  
• Results tables at the shot, scene, and video level: View tagging results at the level you need, with thresholds applied.

• Auto-Evaluator: Quickly assess visual tag performance at the keyframe and shot level, using high-quality ground truth generated on demand. Streamline analysis with a recommended threshold - no manual tuning, for a faster time to value.
  

Powered by a fully rearchitected, multimodal AI platform

One of the most important design principles behind MAP Autumn '25 is flexibility. The platform is fully model-agnostic, meaning customers can leverage best-of-breed foundation models without being locked into a single provider.

As new models emerge—from OpenAI, Deepgram, Google Gemini, Alibaba Cloud’s Qwen, and others—MAP Autumn '25 allows teams to select the right tool for the job. We also enable multimodal fusion, combining specialized models into a chain that outperforms any single model.

In practice, this means customers can continually optimize for accuracy, performance, and cost—future-proofing their content intelligence strategy.

In addition, Coactive enables multi-modal fusion across multimodal signals. Since a point model will often have higher targeted performance against certain tasks as compared to a fully multi-modal model, the ability to use a variety of models (chain of models) to achieve an application goal will be critical for developers. 

‍
Coactive MAP is the most cost-effective way of leveraging your video content

The Coactive platform has been designed to provide customers with the lowest total cost of ownership. Intelligent keyframe sampling cuts compute costs by 6–10x. Cached outputs and tensors eliminate redundant processing, so that you can leverage a single layer of embeddings without having to reprocess.

Why this matters for media companies

For media organizations, the stakes couldn’t be higher. Content libraries are growing, competition for audience attention is intensifying, and advertisers are demanding more personalization and impact. At the same time, brand safety and compliance pressures continue to mount.

With MAP Autumn '25, media companies gain a single, unified platform that can:

• Drive monetization by making it easier to personalize ads and confirm the suitability of brand placements.
  
  
• Streamline operations with automated tagging and search that saves time across teams.
  
  
• Enhance personalization by tagging key characters, products, and cultural moments in real time.
  
  
• Protect trust and safety by automatically filtering inappropriate or unsafe content.
  

Check it out yourself - come to the webinar or talk with our experts

MAP Autumn '25 is now generally available, and we can’t wait to see how you would put it to work.

• Or reach out to our team for a personalized demo. Talk to an expert.
• Learn more in our upcoming webinar, "Unleash your video and image content: Accelerate content workflows with Coactive MAP ‘25"
• Check out the launch blog from the Coactive Head of Engineering, Satya Komala 

Together, let’s unleash the true value of video and image content.

— Cody