兔子API 兔子API
中

home

Model and API documentation

Direct answerThis center contains human-reviewed provider positioning, model-family capabilities, protocol semantics, limits, and official sources; live IDs and prices remain in the marketplace.

Updated · Reviewed

What this documentation answers

This site does not duplicate marketplace IDs, prices, or availability. It answers four durable questions: who provides a capability, what a model family officially supports, how a protocol behaves, and how a workload should be designed. Every capability is derived from the official developer sources listed on the page and is technically paraphrased in Chinese and English.

How the four page types work together

Provider pages distinguish model originators, cloud hosts, and local runtimes so third-party models are not attributed to a channel. Model-family pages explain official product lines, modalities, selection, and migration risk. Protocol pages separate gateway calls from upstream-native semantics. Use-case pages cover architecture, evaluation, and safety.

Boundary with the model marketplace

Exact model IDs, current availability, price, and enabled endpoints are runtime state and belong in the model marketplace. Editorial pages never generate quality claims from SKU names and do not remove durable knowledge merely because a model is temporarily unavailable.

How accuracy is reviewed

Every page records publication, substantive update, review date, and official sources. Changing facts such as versions, context, regions, price, and retirement are included only when verifiable and are placed on a review schedule. If an official source cannot be checked reliably, the limitation is stated rather than filled with speculation.

Beginners should start with the gateway beginner guide to sign in, select a model, fund the account, create a key, make a first call, and inspect its log. Continue to the AI API quickstart when writing a request. For status codes, interrupted streams, or retry behavior, use the API troubleshooting guide. Then read the relevant model-family and protocol pages, and finish in the model marketplace to confirm the exact IDs, prices, and endpoints currently callable on this site. Production still requires evaluation on your own data, latency budget, safety rules, and failure cases.

00

Learning path

From first request to production governance

Beginner · learn the gatewaySign up, choose a model, fund the account, create a key, make a call, and find its log. Intermediate · make it reliableChoose endpoints, streaming, async, media, and billing contracts for a real integration. Expert · govern itEvaluate migrations, control suppliers, protect data, reconcile cost, and operate failure boundaries.
01

Usage guides

Start with the job to be done

Gateway beginner guide: from signup to calls and billing checksA first-time user's path through account access, model selection, funding, API-key creation, first calls, log-based billing checks, and safe support requests.AI API quickstart: from API key to first requestLearn base URLs, API keys, model IDs, and endpoints, then make a first AI API call with curl and read the response.API endpoint and capability matrixMap public gateway paths to capabilities, methods, transports, implementation status, and model-support boundaries without treating route existence as availability.Common AI API error codes and troubleshootingCompare official provider guidance for 400, 401, 402, 403, 404, 408, 409, 413, 422, 424, 429, 499, 5xx, and 529, including retryability, request IDs, and streaming failures.AI API production reliability: timeouts, retries, limits, and SLOsDesign end-to-end deadlines, bounded retries, concurrency protection, circuit breaking, degradation, observability, and load acceptance for AI APIs.SSE streaming and WebSocket realtime connectionsHandle SSE events, in-stream errors, cancellation, proxy buffering, backpressure, and realtime WebSocket sessions from first byte to confirmed completion.Unified async tasks: submit, poll, and retain resultsUse the generic async wrapper and media task APIs while handling queues, progress, terminal states, JSON or binary results, expiration, and duplicate submission.File and media input, output, and lifecycleChoose URL, Base64, data URL, or multipart safely, understand the unsupported Files boundary, and manage temporary media and result URLs.Billing, usage, and cost reconciliationUnderstand runtime prices, groups, pre-consumption, final settlement, refunds, caching, and async tasks, then reconcile request and business cost.Model originators, hosted clouds, aggregators, and local runtimesSeparate model knowledge from the route used to call it across originators, hosted clouds, aggregators, workflows, local runtimes, and account bridges.AI gateway admin runbook: channel rollout, retries, and supplier healthConfigure, test, canary, and roll back channels with least privilege, then diagnose retry, cooldown, catalog, account-balance, and supplier-health failures.Media and object-storage admin runbook: profiles, access, and lifecycleConfigure storage profiles for generated, staging, compliance, and message assets; validate S3 or OSS connectivity, private delivery, lifecycle, cleanup, and recovery.AI API security, privacy, and data governanceCreate verifiable boundaries for API keys, input data, file URLs, prompt injection, tool authorization, logs, and upstream retention.Model evaluation, canarying, and version migrationBuild a representative evaluation set and pinned baseline, then compare offline, canary, roll back, and monitor before a model upgrade or retirement.General text, chat, and reasoning model guideChoose general text models by reasoning depth, context, structured output, tools, latency, and cost.Choosing and calling coding modelsFilter coding models by repository size, tool use, reasoning depth, and latency, then validate on real tasks.Image generation and vision model guideUse multimodal chat for image understanding and dedicated Images endpoints for generation and editing.Video generation and asynchronous API guideVideo generation is usually asynchronous: persist task IDs, limit polling, and validate result media.Speech, transcription, and realtime audio guideSpeech synthesis, transcription, and realtime conversation use distinct endpoints and transports.RAG, embedding, and reranking model guideEvaluate chunking, embedding, retrieval, reranking, and evidence-grounded generation separately.Tool use and AI agent model guideAgents should prioritize tool schemas, long-horizon state, failure recovery, and least privilege.
02

API protocols

Choose the protocol contract

Anthropic Messages API guideCall Claude-protocol models through Messages while distinguishing this site's recommended Bearer token, compatible x-api-key, and direct-version requirements.Audio, transcription, and Realtime APICall speech synthesis, transcription, translation, and WebSocket realtime sessions with correct binary, state, backpressure, and recovery handling.Content Moderation safety classification APIClassify risky text or multimodal input, then integrate calibrated thresholds, human review, audit evidence, and explicit failure policy.Embedding and Rerank retrieval APIBuild an evaluated retrieval pipeline with vector recall, reranking, batching, dimensions, index consistency, score calibration, and RAG authorization.Gemini API guide: Interactions and GenerateContentUnderstand Google's recommended Interactions API and this gateway's implemented GenerateContent multimodal protocol, responses, and migration boundary.Image generation and editing APIGenerate and edit images through the site's OpenAI-compatible endpoints, then handle outputs, failures, safety, and cost in production.Kling-style asynchronous media APICreate Kling video, image, and audio jobs through this site's /kling/v1 compatibility routes while separating direct authentication, job states, and output lifetime.Midjourney task compatibility protocolSubmit, poll, and retain Midjourney jobs through this site's /mj compatibility routes while respecting the boundary with Midjourney's official product and terms.OpenAI-compatible API guideUse one base URL, a Bearer API key, and an exact model name across chat, Responses, and media endpoints.Suno music task-proxy protocolSubmit music, lyrics, and continuation jobs through this site's /suno proxy, poll normalized states, and retain audio safely.Tool calling and structured output APIDeclare function tools through Responses or Chat Completions, execute controlled side effects, and validate strict JSON Schema output.Video generation task APICreate, poll, and download video tasks while handling queued, in_progress, completed, failed, expiry, and production recovery.
03

Model families

Compare stable model families

Anthropic ClaudeAn official-source guide to Claude model tradeoffs, the Messages API, tool use, and migration requirements.DeepSeekAn official-source guide to V4 Flash and V4 Pro thinking controls, streams, tools, compatible APIs, and cost safety.DoubaoAn official-source guide to Doubao general, reasoning, and multimodal capabilities, Ark access, and version checks.FLUX image modelsAn official-source guide to FLUX image generation and editing lines, asynchronous access, and quality tradeoffs.Google GeminiAn official-source guide to Gemini multimodality, model lifecycle, GenerateContent, and model selection.Kling video modelsAn official-source guide to Kling video and image generation, asynchronous jobs, input modes, and launch checks.Meta LlamaAn official-source guide to Llama open-weight capabilities, deployment responsibility, licensing, and access boundaries.Mistral familyAn official-source guide to Mistral general, coding, OCR, voice, and safety models, APIs, and deployment choices.Moonshot KimiAn official-source guide to Kimi long context, coding, vision, thinking modes, and tool-use boundaries.OpenAI GPT and o-seriesAn official-source guide to OpenAI GPT positioning, capability boundaries, Responses API access, and version migration.QwenAn official-source guide to Qwen general, reasoning, coding, and multimodal lines and compatible protocols.Zhipu GLMAn official-source guide to GLM reasoning, vision, image, and agent capabilities and chat-completion access.
04

By model provider

Compare providers and platforms

01.AI YiYi's official platform entry still establishes ownership, but its public documentation is not reliably accessible; models, base URL, and capabilities must be confirmed in the live console.360 AI Model PlatformThe 360 model platform includes both first-party and third-party models; attribution should follow each model's documentation.AWS BedrockAWS-managed platform for foundation models from multiple providers, with unified inference, model discovery, and cloud integrations.Alibaba QwenAlibaba Cloud Model Studio hosts Qwen and third-party models; region, workspace API host, key type, and model protocol must match.Anthropic ClaudeAnthropic Claude provides long-context, coding, and tool capabilities through Messages with typed content blocks and stream events.Azure AI FoundryMicrosoft Foundry, formerly Azure AI Foundry, is Azure's multi-provider model deployment and development platform.Baidu ERNIEBaidu Qianfan provides ERNIE and partner models through a V2 OpenAI-compatible API; bce-v3 keys, preset model IDs, and custom-service API names are distinct.Black Forest LabsFocused on high-quality image generation and editing across realtime previews, production workflows, and fine-grained control.ByteDance DoubaoDoubao models provide text, reasoning, agents, and multimodality through Volcengine Ark; preset model IDs, custom endpoint IDs, and gateway aliases are not interchangeable.ByteDance JimengJimeng developer access is distributed through Volcengine visual services with signed authentication and submit/query jobs; result URLs are not permanent storage.Cloudflare Workers AIA managed inference platform on Cloudflare's edge network with a multi-task model catalog and OpenAI-compatible APIs.CohereCohere provides enterprise chat, Embed, and Rerank; its native V2 APIs differ in path, request schema, and response shape from this site's OpenAI-compatible entry point.DeepSeekDeepSeek currently serves V4 Flash and V4 Pro with switchable thinking through an OpenAI-compatible API; direct and gateway credentials are not interchangeable.Google GeminiGoogle Gemini provides native multimodal models; Google recommends Interactions for new projects while GenerateContent remains supported.HappyHorseHappyHorse can currently be verified only through official Alibaba Cloud Model Studio distribution material; this page does not claim a separately verified model vendor.Jina AIJina AI develops and hosts search-foundation APIs focused on embeddings, reranking, and multimodal retrieval.Kuaishou Kling AIKuaishou Kling Open Platform provides image and video generation APIs for creative content production.Meta LlamaMeta's open-weight model family, available for self-hosting or through partner cloud platforms.MiniMaxMiniMax provides text, speech, and media APIs; pay-as-you-go versus Token Plan keys and OpenAI versus Anthropic base URLs are not interchangeable.Mistral AIMistral AI offers both hosted APIs and open-weight models across chat, code, OCR, speech, and content safety; direct access, this site's compatibility route, and self-hosting have different authentication and licensing boundaries.Moonshot KimiKimi provides long context, reasoning, coding, and tools; mainland China and international accounts, balances, and API keys are isolated.OllamaA model runtime for both local execution and Ollama Cloud, with a model library, native API, and compatibility endpoints.OpenAIOpenAI provides text, reasoning, coding, and multimodal models, and recommends Responses API for new projects.Tencent HunyuanTencent Cloud TokenHub hosts Hunyuan and third-party models; default model IDs, custom online-service IDs, and regional base URLs are not interchangeable.ViduProvides text-, image-, and reference-driven video generation alongside image and sound capabilities.Zhipu GLMZhipu GLM provides OpenAI-compatible chat, multimodal input, and platform tools; client functions and server-side search or MCP have different execution boundaries.iFLYTEK SparkiFLYTEK Spark provides general chat, Spark-X2 reasoning, and function calling; each version's direct path, APIPassword, and model/domain must be used together.xAI GrokA model platform for general generation, reasoning, coding, tool use, realtime voice, and media generation.
Model marketplaceCheck current availability, exact model IDs, and pricing separately from editorial documentation.

FAQ

What is the difference between the documentation center and the model marketplace?

The documentation center explains providers, model families, API protocols, capability boundaries, and usage. The marketplace shows exact model IDs, prices, and endpoints currently callable on this site. Read the guide first, then confirm live state in the marketplace.

Where should a complete beginner start?

Start with the gateway beginner guide and follow account access, marketplace selection, funding, API-key creation, first call, and log lookup in order. Continue to the AI API quickstart when you are ready to write a request.

Why are model IDs and prices not hard-coded in these guides?

Exact IDs, channel availability, prices, and endpoints change with runtime configuration and would become stale quickly. The marketplace owns those dynamic facts, while the documentation preserves more durable protocol and capability knowledge.

How can I tell whether a documentation page is outdated?

Check its substantive update date, review date, official sources, and stated boundaries, then verify the current model ID and endpoint in the marketplace. When upstream or runtime state has changed, use the latest official page and marketplace state.