Twelve Labs · depuis 1. Dezember 2025 (Marengo 3.0, General Availability) · 2× · vu le 14 sept. 2026
100
Momentum
Marengo is a multimodal embedding model by TwelveLabs for video understanding. It analyzes video, image, audio, and text jointly and produces vector embeddings in a shared latent space, enabling any-to-any search (e.g., text-to-video). The current version, Marengo 3.0, was announced in December 2025 and offers reduced embedding dimensions (512 vs. 1024), support for up to four hours of video, and 36 languages. The model is offered both via TwelveLabs' own SaaS API and via Amazon Bedrock (including managed Knowledge Bases).
Historique du momentum
16.06.14.09.
Fonctionnalités
| Deployment (Self-Hosted/Cloud) | Cloud only: TwelveLabs' own SaaS API or Amazon Bedrock (Standard, Priority, Flex service tiers); no self-hosting |
| Throughput/Latency | Marengo 3.0: 50% smaller embeddings (512 vs. 1024 dims), 2x faster indexing vs. predecessor; processes up to 4 hours of video/audio, files up to 6 GB |
| Platform | Available via TwelveLabs SaaS (Search, Embed API) and Amazon Bedrock (incl. Managed Knowledge Bases) |
| Price | Video indexing $0.042/minute; Embed API video $0.042/min, audio $0.008/min, image $0.10/1K, text $0.07/1K; search $4/1K queries; infra $0.0015/min |
| Protocol Compatibility | REST API with x-api-key auth (TwelveLabs SDKs for Python/Node.js); on Bedrock via InvokeModel/StartAsyncInvoke, input/output via Amazon S3 |
| Release Date | Marengo 3.0: December 1, 2025 (GA, announced at AWS re:Invent); Marengo 2.7: December 2024 |
| Supported Models/Providers | Current model version: Marengo 3.0 (previously 2.7, 2.6); Bedrock model IDs: twelvelabs.marengo-embed-2-7-v1:0 / twelvelabs.marengo-embed-3-0-v1:0 |