kriora.com

What It Is

Kriora positions itself as an “infrastructure layer for production AI,” offering a unified API, serverless inference, and on-demand GPU deployments. Its main selling point is an OpenAI-compatible API: in theory, existing applications built on the OpenAI SDK only need to change the base_url to connect to Kriora. The platform’s examples show Chat Completions calls to deepseek/deepseek-r1-0528, suggesting a focus on inference for open-source large language models.

Core Capabilities

Kriora offers two main capabilities. The first is token-based Serverless Inference, designed for quick access to supported models. The second is On Demand Deployments, which lets users deploy open-source models to managed GPUs with one click, including B200, H200, H100, A100, and other instances. It emphasizes a unified API, compute abstraction, reliability, and a smooth path from prototype to production, making it suitable for development teams that do not want to manage GPUs, deployments, or low-level infrastructure themselves.

Pricing and Payment

The pricing model is relatively clear: Serverless inference is billed by input/output tokens with no monthly commitment, though full per-model pricing requires logging in to the platform. GPUs are billed hourly; for example, B200 180GB costs $6.75/hour, while H100 80GB costs $2.97/hour. Payments are processed by Paddle, with support for major credit cards, PayPal, and other methods. Prices are shown in USD, taxes are calculated based on the billing region, and a 14-day refund window is available.

Pros and Cons

The advantages are low migration cost, standardized APIs, and coverage of both inference APIs and GPU deployments, making it a good fit for quickly launching an AI backend. The drawbacks are also clear: the collected materials do not disclose the full model list, token pricing, data retention policy, or whether user data may be used for training. Its terms of service also explicitly state that no SLA is provided, with no commitment to uptime, accuracy, or error-free operation, and that Kriora reserves the right to rate-limit or suspend services at any time. For production-critical workloads, teams should first conduct load testing, disaster recovery planning, and cost validation.

Who It’s For and Access from China

Kriora is best suited for developers and teams familiar with the OpenAI SDK who want access to open-source models or need temporary GPU compute. Chinese-language support, network accessibility from mainland China, and RMB/local payment options are not specified, so mainland Chinese users should test connectivity, latency, and payment availability themselves. If access or compliance becomes a constraint, alternatives to compare include OpenAI, Together AI, Fireworks AI, Replicate, and RunPod; in China, options such as Alibaba Cloud Bailian, Volcano Ark, Tencent Cloud, or SiliconFlow may also be worth considering.