Cohere's Compass Tops Video Retrieval Benchmarks at 0.69
Cohere launched Video Intelligence in Compass on November 16, 2026, ranking first across 11 datasets with an average nDCG@10 score of 0.69 and a 4.4-point lead over rivals.
By Grace Kim
3 min read

What's News
- Cohere launched Video Intelligence in Compass on November 16, 2026, with a design-partner program.
- Compass ranked first overall with an average nDCG@10 score of 0.69 across 11 datasets and 11,133 queries.
- Reported leads included 5.1 points over Gemini Embedding 2 on soccer broadcasts and 4.8 points over Amazon Nova 2 on public benchmarks.
- Video and documents share one permission-aware Compass index, with results carrying timestamps, transcripts and visual descriptions.
- Design partners can deploy in Cohere's managed cloud or their own environment and evaluate against their own footage.
Cohere's Compass scored an average 0.69 across 11 datasets to rank first overall in the company's video-retrieval evaluation, beating seven video embedding models by an average 4.4 points. On November 16, 2026, Cohere used those numbers to launch Video Intelligence in Compass and open a design-partner program for timestamped retrieval of clips, transcripts and visual descriptions alongside documents.
The feature does something fixed-clip libraries can't: Compass returns a bounded video segment chosen at query time. Start and end points depend on the request, not a pre-cut library. Every result ships with a link to the original footage, a timestamp, a transcript and a visual description — so a reviewer can check the evidence against the source.
What does video-in-one-index actually change?
Video is indexed and retrieved inside the same Compass system that handles documents. A single result set can pair a recording segment with a related page, slide or spreadsheet. Permission-aware retrieval limits each user to content they're authorized to see. Organizations can also configure fields, choose models and select deployment options.
The Compass MCP server connects AI agents to the system, letting them analyze slides, charts and demonstrations in addition to spoken material. Cohere frames this as extending enterprise search and agent workflows to visual evidence — while keeping a route back to the original source.
The takeaway: one permission-aware index means an agent can cite a video clip the same way it cites a document.
Where does Cohere think this gets used?
The company's application list ties retrieval to review work:
- Meeting assistants surfacing decisions, forecasts and commitments from recorded meetings, earnings calls or analyst briefings
- Employees finding a specific demonstration or procedure inside training material
- Legal and compliance teams locating statements and disclosures with supporting visual context
- Sports and media teams searching by play, player, camera angle or visible brand for highlights, licensing and sponsor reporting
Each case rests on the same mechanic: find the moment, point the reviewer to that exact point in the footage.
How big are the reported wins?
Cohere's evaluation compared Compass against seven video embedding models across 11 datasets — six public benchmarks and five internal enterprise video datasets. The test ran 11,133 queries and scored results with nDCG@10, which measures how accurately the top 10 results are ranked. Compass finished first overall at 0.69.
The margins varied by domain:
- 4.8 points over Amazon Nova 2, averaged across five open-source public benchmarks
- 5.1 points over Gemini Embedding 2 on soccer broadcasts
- 5.0 points over Gemini Embedding 2 on MultiVENT 2.0 news retrieval
- 2.0 points over Qwen3-VL Embedding 8B on business meetings, lectures and workplace-safety training
Cohere notes the strongest alternative varied by domain — the lead narrows on internal-style enterprise footage, where it was smallest at 2.0 points.
The plain read: Compass leads everywhere it was tested, but the gap is widest on public benchmarks and tightest on workplace video.
Who gets access, and on what terms?
Design partners can deploy Video Intelligence in Cohere's managed cloud or in their own environment. They can evaluate it against their own footage, queries and use cases, and work directly with Cohere's AI search team. Participants are also invited to shape future capabilities based on their organizational needs and experience.
The core enterprise pitch is a shared evidence trail across video and documents: an agent retrieves a decision, demonstration or event and points reviewers to the corresponding moment in the original footage. The design-partner program is what will show whether the reported ranking gains hold up on customer-owned footage and permissioned workloads — the practical test of whether a 0.69 benchmark score turns into retrieval people actually use.
Source: unite.ai
More from Grace Kim
Show full bio
Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,178 articles