Skip to main content
Claude traffic that leans on Anthropic-native features does not have to flatten into the OpenAI shape. POST /api/v1/messages serves the Messages API, untranslated: point the official Anthropic SDK at Belvedir and requests pass through to Anthropic byte-faithfully.
The SDK’s authToken option (or the ANTHROPIC_AUTH_TOKEN env var) works as the key carrier too; Belvedir reads the key from either Authorization: Bearer or x-api-key.

​
What passes through intact

  • Prompt caching: cache_control breakpoints (per-block and top-level) reach Anthropic unmodified, and cache reads bill at the cache-read rate. Verify fidelity yourself by reading usage.cache_read_input_tokens off responses, exactly as you would against Anthropic directly. Cache writes bill at Anthropic’s published write premiums (1.25× for 5-minute, 2× for 1-hour).
  • Thinking: adaptive thinking, display settings, thinking blocks in responses and streams; Belvedir never injects or strips a thinking config on this surface.
  • anthropic-beta headers: forwarded verbatim, except oauth-* betas (a client-auth artifact), which are dropped. Fast mode, compaction, context management and future betas work without waiting on Belvedir; fast-mode calls bill at Anthropic’s published fast rate.
  • Streaming: Anthropic’s native event shape (message_start, content_block_delta, message_delta), byte for byte. A frontend that parses Anthropic SSE keeps working unchanged.
  • The Claude Agent SDK: it honors the same env vars for all its traffic, subagents included. Set ANTHROPIC_BASE_URL=https://platform.belvedir.ai/api and ANTHROPIC_API_KEY=<bv_live_ key> (or ANTHROPIC_AUTH_TOKEN) and its inference goes through Belvedir.

​
Models

This surface is model-pinned by design: pass a real model id. "auto" and routing live on the OpenAI-compatible endpoint. Accepted ids:
  • Claude models, bare (claude-sonnet-5) or prefixed (anthropic/claude-sonnet-5).
  • These hosted open models, which Belvedir serves behind an Anthropic-compatible API: zai-org/GLM-5.2-FP8, moonshotai/Kimi-K2.6, deepseek/deepseek-v4-flash-0731, openai/gpt-oss-120b, google/gemma-4-31B-it, nvidia/Gemma-4-31B-IT-NVFP4.
Any other id returns 400 pointing at the OpenAI-compatible endpoint. A Chinese-lab model on a project with Chinese models off returns 403 (see Data handling).

​
Server tools

Web search and web fetch work here; other Anthropic server tools do not. web_search_<date> and web_fetch_<date> tools pass through and are metered from the response’s usage.server_tool_use: each web search bills a flat 1 cent on top of tokens (Anthropic’s list price of $10 per 1,000 searches); web fetch has no per-call fee. Code execution, computer use, the text editor / bash / memory / tool-search tools, the MCP connector (mcp_servers) and containers bill by container time or outside the usage object, so a request carrying one returns 400 before anything runs. Custom tools (a name plus input_schema, with no type or type: "custom") work everywhere. Forced tool choice is not translated on this surface: Claude Fable 5.1 and Claude Opus 5.5 reject it, so send {"type": "auto"} yourself (see Capabilities and Limits).

​
Cost and metering

Every token is metered from the native usage object, cache writes and fast mode included, at the rates on the Pricing page (Inference → Pricing), attributed to the calling API key. On non-streamed calls, what the call cost rides the x-belvedir-cost response header; the body is never rewritten. Streamed calls carry no cost header: read spend off the usage page. A streamed response the client aborts is billed on an estimate of what was streamed.

​
Errors and limits

Errors come back in Anthropic’s error envelope, so the SDK’s typed errors work; upstream errors pass through verbatim. Request bodies cap at 20 MB (413 beyond it). The key auth, the 402 spend gate and the rate limit (25 requests per second per API key, burst 300, adjustable per organization) are the same as on chat completions.

​
Message Batches

With the Anthropic client pointed at https://platform.belvedir.ai/api, client.messages.batches.create / retrieve / results / cancel / list / delete run against /api/v1/messages/batches, which speaks Anthropic’s Message Batches wire shapes over Belvedir’s batch tier. It is documented with the other batch endpoints on Batches.

​
What no router can carry

Anthropic’s Managed Agents / Agent API sessions run their inference inside Anthropic’s own orchestration; there is no base URL to override, for Belvedir or anyone else. Instrument those workloads with the Belvedir SDK for observability instead.