兔子API 兔子API
中

provider

Cloudflare Workers AI

Direct answerA managed inference platform on Cloudflare's edge network with a multi-task model catalog and OpenAI-compatible APIs.

Updated · Reviewed

Separate the hosting platform from model authors

Cloudflare Workers AI runs models from many originators on Cloudflare's network. Cloudflare is the host and billing platform; catalog models from Meta, BAAI, Mistral, BFL, and others are not Cloudflare-authored. Full IDs commonly use @cf/<author>/<model>, and the exact catalog record defines license, modality, fields, and lifecycle.

Native and OpenAI-compatible paths differ

Direct native REST uses /client/v4/accounts/{account_id}/ai/run/@cf/<author>/<model>, with the model in the URL. The OpenAI-compatible base is /client/v4/accounts/{account_id}/ai/v1, followed by Chat Completions, Responses, or Embeddings with the full ID in JSON. The listed /v1/* routes are adapters under this site's base URL, use this site's Bearer token, and do not contain a Cloudflare Account ID.

Compatibility does not unify the entire catalog

Native tasks span text, embeddings, images, speech, and other schemas. OpenAI compatibility applies only to the resources and models Cloudflare documents. Tools, vision, streams, context, and errors can still differ by model under one path. Run field-level contract tests against the exact @cf/... ID and monitor deprecation notices.

Batch is a separate asynchronous capability

Workers AI Batch queues a collection of requests, returning queued and request_id at creation and providing results later. It fits offline summarization or embeddings. Total payload must remain below the official limit, and only catalog models marked for Batch are eligible. This site's current api_paths do not list Batch, so the Cloudflare-native Batch route must not be presented as a gateway endpoint.

Neurons, 429, and cost

Cloudflare uses Neurons as a cross-model GPU-compute unit in its backend. The catalog converts tokens, images, or other units into Neurons; free allocation and paid rates can change, so use current Pricing. A 429 can mean exhausted daily free allocation or unavailable capacity. Inspect Cloudflare's error code before upgrading, throttling, or bounded retry. Set per-model concurrency, daily budgets, and usage alerts, and grant API tokens only the required Workers AI Read/Edit permissions.

Reviewed provider catalog

Latest models by capability

These are provider-catalog models, not a promise of availability on this site. Confirm callable IDs, endpoints, and pricing in the model marketplace.

Text, reasoning, and code

Image generation and editing

  • @cf/black-forest-labs/flux-2-klein-9b Open weights

    Black Forest Labs' open-weight FLUX.2 image model hosted by Cloudflare.

    Origin: Black Forest Labs Release date not stated in the cited official source
    Official model catalog ↗ Model details →

Speech and audio

Embeddings, retrieval, and reranking

  • @cf/qwen/qwen3-embedding-0.6b Open weights

    Qwen's lightweight Qwen 3 text embedding model hosted by Cloudflare.

    Origin: Qwen Release date not stated in the cited official source
    Official model catalog ↗ Model details →
Check live availability in the model marketplace

Use cases

  • Managed edge inference
  • Multi-task model catalog
  • OpenAI-compatible access

API protocols

  • /v1/chat/completions
  • /v1/responses
  • /v1/embeddings

FAQ

How do Workers AI native REST and OpenAI-compatible paths differ?

Native REST places the model in /ai/run/@cf/author/model. The compatible API uses account-level /ai/v1 as its base URL and sends the full ID in the JSON model field.

Does @cf/meta/... mean Cloudflare developed the model?

No. @cf is Cloudflare's hosted namespace, while the following author segment identifies Meta, BAAI, or another originator. Read the exact catalog entry for license and capability.

Does Workers AI Batch return every result immediately?

No. It first returns queued and a request_id; results are retrieved later. The total payload must stay below the documented limit and the model must be marked Batch-capable.

Are Cloudflare Neurons the same as model tokens?

No. Neurons are Cloudflare's common GPU-compute and backend billing unit. Different models convert text tokens, images, audio, or other work into different Neuron amounts.

Does every Workers AI 429 mean requests arrived too quickly?

No. Official error codes distinguish exhausted free Neuron allocation from data-center capacity. Inspect the code before choosing an upgrade, throttling, or bounded retry.

  • OpenAI-compatible API guideThis site's OpenAI-compatible routes use a Bearer token; Responses, Chat Completions, and media endpoints have distinct objects and streaming semantics and must match the target model.
  • RAG, embedding, and reranking model guideRAG combines chunking, embeddings, retrieval, reranking, grounded generation, and evaluation; completeness comes from citable evidence and no-answer handling, not a larger model list.
  • Tool use and AI agent model guideAn agent is a controlled loop, not merely tool calling: structured schemas, least privilege, state, idempotency, timeouts, approval, audit, and replayable evaluation.
  • AI API quickstart: from API key to first requestAn AI API call needs a matching base URL, API key, model ID, and endpoint; start with minimal curl, then add error handling, cost controls, and key security.
  • Common AI API error codes and troubleshootingHandle AI API failures by parsing the HTTP status, provider code, error type, message, retry hints, and request ID together. Fix request, authentication, billing, and hard-quota causes first; retry only transient throttling, network failures, and selected 5xx errors with strict bounds.

Official sources

  1. Cloudflare Workers AI Models Official
  2. Workers AI OpenAI Compatibility Official
  3. Workers AI REST API Quickstart Official
  4. Workers AI Asynchronous Batch API Official
  5. Workers AI Pricing and Neurons Official
  6. Workers AI Errors Official