Reka Edge

Reka Edge is a multimodal model from Reka AI, listed on api14 under the model id rekaai/reka-edge. It is a 7B vision-language model built for efficiency, which matters when you want multimodal understanding without running a very large model. The provider describes it as optimized specifically for image understanding, and it sits in the text category of the catalogue, meaning text generation is its primary output.

On the input side, Reka Edge accepts images, video and text, and it produces text. That makes it a fit for workloads where the prompt combines visual media with written instructions and the answer comes back as prose, structured text or tool calls. The context window is 16,384 tokens across input and output, and the maximum output is 14,745 tokens, so you can request fairly long generations while still leaving room for the prompt. Supported request parameters include temperature, top_p, top_k, frequency_penalty, presence_penalty, seed, stop, max_tokens, logprobs and top_logprobs, so you have the usual sampling controls along with token-level probability inspection.

Because it handles image and video input alongside text, the obvious use cases are ones where the model has to read visual content rather than just describe it in general terms. Screenshot and document image understanding, extracting or summarizing what is in a picture, answering questions about a video clip, and building assistants that take a mixed text and media prompt all follow from the stated capabilities. The structured_outputs parameter means you can push responses toward machine-readable shapes, and tools plus tool_choice let you wire the model into function calling flows, so it can act as the reasoning step in an agent that operates on visual inputs.

Calling it through api14 works the same way as any other model on an OpenAI-compatible endpoint. You send a chat completion request with the model field set to rekaai/reka-edge, include your messages with text and image or video content as needed, and pass any of the supported parameters in the request body. No separate SDK or translation layer is required, so existing OpenAI client code needs only the model id changed and a media payload added where relevant.

The main constraint to keep in mind is the 16,384 token context. If your inputs involve long videos, large documents or extended back-and-forth history, that budget will fill quickly, and a model with a larger context may suit the task better. Reka Edge is most useful when you want a compact multimodal model that handles image and video understanding with structured output and tool support, rather than a text-only model or one aimed at very long contexts.