Provider request: XiuRouter with public catalog and source-linked pricing

XiuRouter is a usage-based API gateway for models from multiple labs. I am submitting on behalf of its operator, XiuLab Inc, and used AI assistance to prepare this request.

Please consider XiuRouter for the provider catalog. The immediate user benefit is that OpenCode users currently have to define provider.xiurouter and enumerate model IDs manually; our current integration guide explicitly documents that gap.

Verified Provider Fields

The proposed provider fields below come from the published OpenCode and Vercel AI SDK guides, not an inferred endpoint:

name = "XiuRouter"
npm = "@ai-sdk/openai-compatible"
env = ["XIUROUTER_API_KEY"]
api = "https://router-api.xiu.ai/v1"
doc = "https://docs.xiu.ai/router/"

• OpenCode integration guide: custom provider using Chat Completions.
• Vercel AI SDK integration guide: uses createOpenAICompatible with the same base URL and environment variable.
• API compatibility.
• Models, service groups and billing.

Provider Mark

The official monochrome mark is available in the published XiuAI package:
https://cdn.jsdelivr.net/npm/@xiuai/design-tokens@0.2.1/dist/assets/logo-mark-mono.svg

Its five path definitions match the mark currently served on the XiuRouter pricing page. A currentColor SVG adaptation with a 0 0 64 64 viewBox and no fixed width/height is prepared for the eventual PR; light and dark render previews have been checked. The four visible mark paths are unchanged, and the center cutout is expressed as a clipping path so no fixed display colors are needed.

Public Catalog And Price Evidence

• Human-readable pricing: https://router.xiu.ai/en/pricing
• Anonymous catalog and pricing inputs: https://router.xiu.ai/api/pricing
• Authenticated, key-scoped model discovery: GET https://router-api.xiu.ai/v1/models

Observed September 19, 2026, Asia/Taipei: the anonymous catalog contains claude-opus-5; the public max service tier's ordinary displayed rate is USD 1.15 per million input tokens and USD 5.75 per million output tokens. These are XiuRouter's rates, separate from the page's reference prices of USD 5 and USD 25. Cache rates are separate, and the human-readable page rounds their display. Fast or account-specific pricing must not be silently substituted for the ordinary public rate.

The catalog exposes exact model IDs and supported_endpoint_types. Protocol availability is not a claim that models from every named lab are currently offered. For example, a Gemini protocol route does not establish Gemini-family model availability.

Remaining Catalog Requirements

I read the repository's base_model, override-only, logo, cost-tier and reasoning-options guidance. This is a provider request with source evidence, not a claim that a ready-to-merge catalog entry or runtime test is complete.

The public pricing feed does not establish all of the host-specific context/output limits, reasoning controls, attachment, tool-call and structured-output behavior required for resolved model entries. Our own guides say these need to be checked for each model and key. I have not sent inference requests as part of this submission and do not want to present inherited lab capabilities as verified gateway behavior.

Would you consider this provider once that per-model evidence accompanies the catalog PR? For an initial implementation, should the provider remain scoped to the documented OpenAI-compatible Chat Completions surface, with other protocol surfaces handled separately?

The provider TOML and logo adaptation are prepared; a catalog PR still needs verified per-model metadata and bun validate. Please do not import all models using guessed capability flags from this issue alone.

Duplicate Check

Before opening this request, I checked the current public https://models.dev/api.json, repository code for router.xiu.ai, and all-state issues/PRs for both XiuRouter and the domain. No existing provider or request was found.

Start by reading the repository's base_model, override-only, logo, cost-tier and reasoning-options guidance, then inspect the current catalog and provider entries. Verify per-model metadata against the linked XiuRouter sources, keep the initial scope to the documented OpenAI-compatible Chat Completions surface, and finish with the provider TOML, logo adaptation, catalog evidence, and a passing bun validate.