# Supported Translation & OCR Providers

> Beyond Translate for macOS supports zero-key translation and OCR out of the box (macOS System Translation on macOS 26+, downloadable Hy-MT2 and Gemma 3 local models, built-in macOS OCR, and offline PaddleOCR), plus BYOK connections for OpenAI, Claude, Gemini, DeepSeek, Qwen, Kimi, Zhipu, Doubao, SiliconFlow, and any OpenAI-compatible server such as Ollama, OpenRouter, or self-hosted vLLM/SGLang.

- Last updated: 2026-11-19
- Canonical URL: https://beyondtranslate.com/providers

Use macOS system translation and on-device models without an API key, or connect your own cloud AI accounts and OpenAI-compatible endpoints.

## Built-in & On-Device Engines (No API Key)

Run translation and screenshot text recognition directly on your Mac without creating a cloud provider account.

| Provider / Engine | Capabilities | Credentials | Notes & Requirements |
| --- | --- | --- | --- |
| macOS System Translation | Translation | No API key | Uses Apple’s system translation framework. Requires macOS 26 or later and downloaded language packs in System Settings. |
| On-device Hy-MT2 | Translation (streaming & plain text) | No API key | Managed inside Settings → Providers → Local models. 1.8B Q4 (~1.13 GB, ≥4 GiB RAM) and 7B Q4 (~4.62 GB, ≥8 GiB RAM) with resumable SHA256-verified downloads. |
| On-device Gemma 3 | Translation | No API key | Selectable from Local models → Manage downloads; inference runs locally after downloading the model files. |
| Built-in macOS OCR | Screenshot & clipboard image OCR | No API key | Native macOS screen text recognition; requires Screen Recording permission. |
| Offline PaddleOCR (PP-OCRv6 Small) | Multilingual offline OCR | No API key | Downloadable offline OCR engine managed in provider settings; can be set as the default OCR service for screenshots and dropped images. |

## Featured Cloud AI Providers (BYOK)

Connect directly using your own API key (Bring Your Own Key). Optional custom Base URLs are supported.

| Provider / Engine | Capabilities | Credentials | Notes & Requirements |
| --- | --- | --- | --- |
| DeepSeek | Translation, OCR, TTS | API Key (+ optional Base URL) | Supported in first-run onboarding and Settings; also supports DeepSeek OCR and speech synthesis. |
| OpenAI | Translation, TTS | API Key (+ optional Base URL) | Uses OpenAI Chat Completions compatible endpoints. |
| Anthropic (Claude) | Translation | API Key (+ optional Base URL) | Connects directly to Anthropic’s Claude API. |
| Google Gemini | Translation | API Key (+ optional Base URL) | Uses Gemini’s OpenAI-compatible endpoint. |
| Qwen (Alibaba Cloud DashScope) | Translation | API Key (+ optional Base URL) | Connects via DashScope OpenAI-compatible mode. |
| Kimi (Moonshot AI) | Translation | API Key (+ optional Base URL) | Connects directly to Moonshot Kimi API. |
| Zhipu AI (GLM) | Translation | API Key (+ optional Base URL) | Connects to Zhipu BigModel API. |
| Doubao (Volcengine Ark) | Translation | API Key (+ optional Base URL) | Uses Volcengine Ark OpenAI-compatible endpoints. |
| SiliconFlow | Translation | API Key (+ optional Base URL) | Access hosted open-weight models through SiliconFlow API. |

## OpenAI-Compatible & Self-Hosted Endpoints

Connect any third-party aggregator or local/private inference server using the OpenAI Compatible provider.

| Provider / Engine | Capabilities | Credentials | Notes & Requirements |
| --- | --- | --- | --- |
| OpenAI Compatible | Translation | Base URL (API Key optional) | Works with self-hosted vLLM or SGLang (e.g. tencent/Hy-MT2-30B-A3B), local Ollama servers, OpenRouter, Groq, Mistral, or internal enterprise proxies. |

## Availability Note on Traditional Translation APIs

Traditional machine-translation APIs such as Baidu Translate, DeepL API, and Google Cloud Translation, as well as standalone dictionary and read-aloud panels, are currently not exposed in the native macOS interface. Check the active service list inside your installed version.

## Provider Questions

### Can I use Beyond Translate completely without an API key?

Yes. macOS System Translation (on macOS 26+), built-in macOS OCR, offline PaddleOCR, and on-device Hy-MT2 or Gemma 3 models all work without any cloud API key.

### What Mac specs are needed for local Hy-MT2 translation?

Hy-MT2 1.8B Q4 downloads approximately 1.13 GB and requires at least 4 GiB of system RAM. Hy-MT2 7B Q4 downloads approximately 4.62 GB and requires at least 8 GiB of system RAM. All model downloads resume after interruption and verify against pinned SHA256 hashes.

### How do I connect Ollama, OpenRouter, or self-hosted vLLM/SGLang?

Add a service using the OpenAI Compatible provider in Settings → Services and enter your server’s Base URL (and API key if required). For self-hosted Hy-MT2 models like tencent/Hy-MT2-30B-A3B on vLLM or SGLang, leave the custom system prompt blank so the built-in Hy-MT2 translation instructions apply.

### Can I compare translations from multiple models at once?

Yes. You configure one default translation service for your primary answer and can enable additional services for side-by-side comparison inside the translator.

### Who receives the text or screenshots I translate?

When you use on-device local models or offline PaddleOCR, text and images stay on your Mac. When you choose a cloud AI provider, submitted text or images are sent directly to that provider under their privacy policy and billing terms—Beyond Translate does not require an account.
