Reka Cloud

What it does

Reka Cloud takes in text, image, video, and audio requests and returns model outputs through a single OpenAI-compatible endpoint. Developers send a standard request shape and can switch between models such as DeepSeek V4 Flash, Qwen 3.8 27B, and GLM 5.3 Flash without rewriting code. Streaming responses are supported.

Teams use it in three main ways: calling open-source models serverless, running training and inference on dedicated GPU capacity including H100, H200, and Blackwell, and customizing models with fine-tuning, LoRA, or reinforcement learning on their own data. Claru, Reka's platform for collecting, annotating, and generating training data, supports the data-preparation step. Reka also describes distillation and 3.5-bit quantization as inference optimizations.

It suits engineering teams building production AI features that need multimodal models, custom training, or dedicated GPU infrastructure. Access begins through the developer portal, and dedicated compute involves reserving capacity. Concrete rates are not listed, so buyers should contact Reka to confirm cost. Teams wanting only a chat interface may find it a poor fit.

What it can do

• Serverless inference for leading open-source models with automatic scaling and pay-per-use billing
• Dedicated GPU compute on H100, H200, and Blackwell hardware with reserved capacity and root control
• OpenAI-compatible API with one request shape across language, image, video, and audio
• Fine-tuning with parameter-efficient methods such as LoRA on proprietary data
• Reinforcement learning for post-training models against specific tasks and outcomes
• Claru platform for collecting, annotating, and generating training datasets

Before you choose

• Pricing for serverless, dedicated compute, and custom training is not listed; confirm costs with Reka before committing.
• Dedicated compute and customization require engineering effort and reserved capacity rather than a self-serve chat experience.
• Models are shown by example; confirm the current model catalog before committing to a specific model.

Listed under

• Coding
• Sandboxes and infrastructure

Get started

The page links to the developer portal to begin; dedicated compute requires reserving capacity.

1. Open the developer portal at developer.reka.ai and create an account.
2. Obtain an API key and send requests to the OpenAI-compatible endpoint, selecting a model such as DeepSeek V4 Flash or Qwen 3.8 27B.
3. For custom work, prepare proprietary training data, then choose fine-tuning or reinforcement learning, or contact the team to reserve dedicated GPU capacity.

Serverless inference is described as pay-only-for-what-you-use. Dedicated compute is purchased by reserving capacity, and custom training costs should be confirmed with the Reka team.

We couldn’t find published prices for Reka Cloud.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Reka Cloud ↗reka.ai
• Reka ↗reka.ai

Company website

reka.ai: product, pricing and policy pages

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.