Introducing Clef-omni with full multimodality, plus a faster Clef and a cheaper Clef-flash

Introduction

Cloudflare’s AI team is expanding its open‑weight decision‑model lineup. The latest release, Clef‑omni, adds true multimodality—accepting audio, video, image, and text inputs in a single model. At the same time, the existing Clef model has been optimized for speed, and Clef‑flash has been repriced to be more affordable than competing offerings. These updates illustrate Cloudflare’s rapid‑iteration approach and its commitment to making powerful AI tools accessible to developers.

Clef‑omni: One Model, All Modalities

Traditionally, decision models have been text‑centric, requiring separate pipelines for speech‑to‑text, image analysis, or video frame extraction. Clef‑omni removes that friction:

• Audio support – Accepts common formats such as WAV and MP3.
• Video support – Handles MP4 and WebM files, processing visual and auditory streams together.
• Image and text – Continues to accept the image and text inputs that earlier Clef versions supported.

By consolidating these capabilities, developers can call a single endpoint to make decisions that span any combination of modalities. This simplifies architecture, reduces latency, and opens new use‑cases such as real‑time content moderation that considers spoken words and visual cues simultaneously.

Faster Clef: Performance Gains

The original Clef model, which already supported images and video frame arrays, has been tuned for speed. Optimizations in the inference pipeline and model weight layout deliver lower latency without sacrificing accuracy. For workloads that require rapid turn‑around—such as interactive chatbots or on‑device inference—this faster version provides a noticeable improvement in user experience.

Cheaper Clef‑flash: Lowering the Barrier to Entry

Clef‑flash, Cloudflare’s lightweight decision model designed for cost‑sensitive scenarios, has been repriced to sit below the price point of comparable offerings. The price reduction makes it attractive for high‑volume or edge‑deployed applications where budget constraints previously limited adoption. Despite the lower cost, the model retains the core capabilities that make it suitable for quick classification and rule‑based decisions.

Practical Takeaways

• Simplify multimodal pipelines – Use Clef‑omni to replace multiple specialized models with a single API call.
• Boost responsiveness – Switch to the faster Clef for latency‑critical applications.
• Control spend – Leverage the cheaper Clef‑flash for large‑scale or edge deployments.
• Stay updated – All three models are released with open‑weight checkpoints on Hugging Face, and detailed integration guides are available in Cloudflare’s developer documentation.

Closing Thoughts

Clef‑omni marks a step toward truly multimodal AI that can perceive the world as humans do—through sound, sight, and text—all within one decision model. Coupled with performance and pricing improvements to the existing Clef family, Cloudflare is positioning its AI platform as a flexible, cost‑effective solution for developers building next‑generation applications. Keep an eye on future releases as the company continues to iterate rapidly, turning ambitious ideas into production‑ready tools.