Tensormesh
Tensormesh serves open-weight models on a KV-cached vLLM fleet and bills cached input at $0. Anyray routes to it as a first-class provider and prices the cached share at zero.
| Property | Value |
|---|---|
| Provider id | tensormesh |
| Gateway speaks | OpenAI /v1/chat/completions; Anthropic /v1/messages is translated |
| Credential | A Tensormesh inference API key |
| Base URL | https://serverless.tensormesh.ai/v1, built in |
| Model ids | Hub names as served, such as Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8 |
Set it up
In Tensormesh, go to Deploy, then Serverless.
Console → Providers → the tensormesh key, or set ANYRAY_PROVIDER_KEY_TENSORMESH
(provider keys).
On the Routing page pick provider tensormesh and a served model id, or send
"provider": "tensormesh" in x-anyray-config (routing).
The served catalog needs no key:
curl https://serverless.tensormesh.ai/v1/models
Tensormesh serves open-weight models only. Route a Claude Code seat here for a Qwen or Kimi coding model. Claude itself stays on Anthropic, where it keeps Anthropic's own prompt cache.
What Anyray meters
Tensormesh reports cached input as prompt_tokens_details.cached_tokens.
| Case | How it prices |
|---|---|
| A model whose price card lists cached tokens at $0 | The cached share prices at $0 |
| Any other model | The cached share prices at the list input rate |
| A served model not yet in Anyray's price table | The provider floor, the cheapest Tensormesh rate, never the house rate |
Streams ask Tensormesh for the final usage frame, so a streamed turn records the same split as a blocking one.