Avian.io

Avian.io is a pay per token AI inference API for open weight large language models. It serves models from DeepSeek, Moonshot's Kimi line, Zhipu's GLM family, MiniMax and Xiaomi's MiMo through an OpenAI compatible endpoint, so an application already written against the OpenAI SDK can switch by changing the base URL and API key. The service is built for developers and teams that want fast, inexpensive inference for coding tools and production applications. Models run on NVIDIA B200 GPUs with speculative decoding, with no rate limits and always warm instances, and a single key covers every hosted model. Vision analysis, web search, a web reader and native tool calling are available across the lineup, and Avian documents use with more than twenty coding tools including Claude Code, Cursor, Cline, Windsurf, Kilo Code and Aider. Infrastructure is hosted on Microsoft Azure with SOC 2 approved controls, GDPR and CCPA compliance and zero data retention after requests. For guaranteed capacity, dedicated deployments on NVIDIA H200 or H100 GPUs can be arranged. Pricing is published per million tokens and drawn from prepaid credits, with no subscription or minimum spend. DeepSeek V4 Flash is $0.0805 per million input tokens and $0.161 per million output tokens with a one million token context window. DeepSeek V4 Pro is $1.305 input and $2.61 output, MiniMax M2.5 is $0.27 input and $1.08 output, and Kimi K2.5 is $0.45 input and $2.20 output. Cached input is billed at lower rates, credits can be topped up in amounts from $25, and dedicated deployments are priced through sales. The catalog changes frequently as new open models ship, and older versions such as DeepSeek V3.2 are marked as legacy, so model availability is worth checking before building against a specific version.