# Zopyr AI — Unified Multi-Provider AI Inference Gateway

> Zopyr AI routes OpenAI, Anthropic, Google Gemini, DeepSeek, xAI Grok, Qwen, Mistral, Kimi, and Meta Llama models through one unified, high-availability API with sub-25ms gateway overhead, automatic multi-provider failover, and zero-data-retention privacy by default.

- Website: https://zopyr.com
- Persian Website: https://zopyr.com/fa
- Gateway Base URL: `https://api.zopyr.com/v1`
- Full LLM Technical Reference: https://zopyr.com/llms-full.txt
- OpenAPI 3.1 Specification: https://api.zopyr.com/docs/gateway.json

## Core Documentation & Platform Pages

- [Models Catalog](https://zopyr.com/models): Browse 300+ AI models with real-time pricing, context window, and latency metrics.
- [Model Comparison Tool](https://zopyr.com/models/compare): Side-by-side comparison of context limits, token pricing, and capabilities.
- [Live Benchmarks](https://zopyr.com/benchmarks): Empirical quality, TTFT latency, and throughput leaderboards.
- [Pricing](https://zopyr.com/pricing): Pay-as-you-go token pricing with zero monthly subscriptions.
- [Developer Documentation](https://zopyr.com/docs): Quickstart guides for OpenAI SDK, Responses API, Anthropic SDK, Claude Code, Embeddings, Files, and Realtime WebSocket.
- [Engineering Blog](https://zopyr.com/blog): Deep-dive benchmarks, distributed systems architecture, and FinOps guides.
- [System Status](https://zopyr.com/status): Real-time gateway uptime and regional latency telemetry.

## Supported Gateway Endpoints (`https://api.zopyr.com/v1`)

- `POST /v1/chat/completions` — OpenAI Chat Completions (JSON & SSE `stream: true`, tools, structured outputs, reasoning, multimodal vision & PDF files, `models` fallback array, `provider.sort`).
- `POST /v1/responses` — OpenAI 2026 Responses API (JSON & SSE `response.output_text.delta`).
- `POST /v1/messages` — Anthropic Messages API compatible with `@anthropic-ai/sdk` and Claude Code (`x-api-key` or `Authorization: Bearer`).
- `POST /v1/embeddings` — Vector embeddings (`float` and `base64`, configurable `dimensions`).
- `POST /v1/completions` — Legacy text completions (JSON & SSE).
- `POST /v1/files`, `GET /v1/files`, `GET /v1/files/:id/content`, `DELETE /v1/files/:id` — Organization-scoped encrypted file uploads with `file_id` injection in chat/responses.
- `POST /v1/images/generations`, `POST /v1/audio/transcriptions`, `POST /v1/audio/speech` — Multimodal media endpoints.
- `GET /v1/realtime` (WebSocket `wss://api.zopyr.com/v1/realtime`) — Stateful OpenAI Realtime WebSocket protocol.
- `GET /v1/models`, `GET /v1/credits`, `GET /v1/key`, `GET /v1/generation?id=gen-...` — Catalog, wallet balance, and per-request cost telemetry.

## Featured 2026 Frontier Models

- [openai/gpt-6.1-sol-pro](https://zopyr.com/models/openai/gpt-6.1-sol-pro): OpenAI flagship reasoning & autonomous software engineering model. (Context: 1,000,000 tokens, Input: $2.00/1M, Output: $8.00/1M)
- [anthropic/claude-opus-5.5](https://zopyr.com/models/anthropic/claude-opus-5.5): Anthropic flagship autonomous coding, deep research, and multi-agent orchestrator. (Context: 1,000,000 tokens, Input: $5.00/1M, Output: $25.00/1M)
- [anthropic/claude-sonnet-5.5](https://zopyr.com/models/anthropic/claude-sonnet-5.5): High-speed production coding, tool-use, and computer-use workhorse. (Context: 1,000,000 tokens, Input: $2.50/1M, Output: $12.50/1M)
- [google/gemini-4-argon](https://zopyr.com/models/google/gemini-4-argon): Google 2M-context native multimodal & mathematical reasoning flagship. (Context: 2,000,000 tokens, Input: $1.25/1M, Output: $5.00/1M)
- [google/gemini-3.8-flash](https://zopyr.com/models/google/gemini-3.8-flash): Ultra-low-latency high-throughput multimodal flash model. (Context: 1,000,000 tokens, Input: $0.10/1M, Output: $0.40/1M)
- [deepseek/deepseek-v4.1-flash](https://zopyr.com/models/deepseek/deepseek-v4.1-flash): High-efficiency MoE reasoning, math, and code generation model. (Context: 256,000 tokens, Input: $0.14/1M, Output: $0.28/1M)
- [x-ai/grok-4.7](https://zopyr.com/models/x-ai/grok-4.7): xAI frontier reasoning, real-time synthesis, and technical analysis model. (Context: 512,000 tokens, Input: $2.00/1M, Output: $10.00/1M)
- [qwen/qwen-3.8-max-prime](https://zopyr.com/models/qwen/qwen-3.8-max-prime): Multilingual enterprise reasoning, Persian/English fluency, and agentic tool use. (Context: 256,000 tokens, Input: $0.40/1M, Output: $1.20/1M)
- [mistralai/mistral-large-4](https://zopyr.com/models/mistralai/mistral-large-4): Mistral flagship structured-output, code, and multilingual model. (Context: 256,000 tokens, Input: $1.50/1M, Output: $4.50/1M)
- [moonshotai/kimi-k3](https://zopyr.com/models/moonshotai/kimi-k3): Long-horizon 2M-context document synthesis and research model. (Context: 2,000,000 tokens, Input: $0.50/1M, Output: $2.00/1M)
- [meta-llama/llama-4-maverick](https://zopyr.com/models/meta-llama/llama-4-maverick): Open-weights multimodal Mixture-of-Experts frontier model. (Context: 512,000 tokens, Input: $0.20/1M, Output: $0.60/1M)

## Engineering Articles

- [GPT-6.1 Sol Pro vs Claude Opus 5.5 vs Gemini 4 Argon: 2026 Production Benchmark](https://zopyr.com/blog/frontier-models-benchmark-2026): An empirical evaluation of reasoning accuracy, time-to-first-token (TTFT), long-context recall, and token economics across the three flagship 2026 frontier models.
- [Inside Zopyr’s Zero-Postgres Hot Path: Sub-25ms Failover & Distributed Circuit Breaking](https://zopyr.com/blog/multi-provider-failover-architecture): How we engineered our inference gateway to authenticate keys, reserve nano-USD holds, evaluate sliding-window circuit breakers, and fail over across upstream providers without touching Postgres.
- [Unified SDK Guide: OpenAI Responses API, Anthropic Messages & Claude Code on One Gateway](https://zopyr.com/blog/openai-responses-and-anthropic-messages-guide): Connect the modern OpenAI Responses SDK (/v1/responses), Anthropic SDK (/v1/messages), and CLI agents like Claude Code to Zopyr using a single API key.
- [Cutting LLM Inference Spend by 68% with Prompt Caching and Price-Sorted Routing](https://zopyr.com/blog/token-cost-optimization-prompt-caching): Practical architectural patterns for structuring system prompts, leveraging provider.sort="price", and cascading from flash models to frontier reasoning models.
- [Building Low-Latency Realtime WebSocket & Multimodal RAG Pipelines on Zopyr](https://zopyr.com/blog/realtime-websocket-and-multimodal-rag): Step-by-step tutorial on combining /v1/files, /v1/embeddings, and persistent /v1/realtime WebSocket sessions for interactive multimodal assistants.
- [Zero-Data-Retention by Default: Architectural Privacy and Encrypted Auditability](https://zopyr.com/blog/zero-data-retention-enterprise-ai-security): How Zopyr protects proprietary source code and customer data through metadata-only request logging, AES-256-GCM envelope encryption, and data_collection="deny" routing.