Valar Inference

As described on AWS Marketplace.

Valar is the inference platform for agentic and async workloads. We run leading open-weight models on an optimized stack across GPUs and alternative accelerators, delivering 5-20x lower cost than frontier APIs with better reliability and throughput.

That cost advantage lets teams run longer and more capable background agents - larger models, deeper reasoning loops, broader search, more retries, and stronger validation on every task. The result is smarter long-running AI workflows, higher throughput, better reliability, and dramatically better intelligence per dollar.

Highlighted by the publisher on AWS Marketplace.

Run popular open-weight models - Kimi K, GLM, DeepSeek, MiniMax M3, Qwen3.5, gpt-oss and more, behind one OpenAI-compatible API, optimized for the economics of long-running agent and async work rather than the speed of a single call.

Build agentic and compound systems on the primitives you already use: tool calls, structured (JSON-schema) outputs, background/async mode, webhooks, and idempotency across the Responses and Chat Completions APIs.

Pick a completion window per request and trade wall-clock time for price: Standard runs about 50% below competitor serverless rates, Flex about 65%, with automatic prefix caching on every request.

As stated by the publisher on AWS Marketplace.

All fees are non-cancellable and non-refundable except as required by law.