AI Model Setup

Configure your AI provider for translation generation. Translation tasks are structured and instruction-following — budget models handle them well, so there's no need to reach for flagship pricing.

Supported Providers

1. LM Studio (Recommended for Local Development)

LM Studio allows you to run AI models locally without cloud dependencies. This is ideal for development, testing, and privacy-sensitive work.

Download from lmstudio.ai, load a model (recommended: Qwen 3 4B), and start the local server.

Configuration

import { createOpenAICompatible } from "@ai-sdk/openai-compatible";

const lmstudio = createOpenAICompatible({
  name: "lmstudio",
  baseURL: "http://127.0.0.1:1234/v1",
});

export default {
  model: lmstudio("qwen3-4b"),
};

2. OpenAI

Get an API key from platform.openai.com and set OPENAI_API_KEY in your environment.

import { openai } from "@ai-sdk/openai";

export default {
  model: openai("gpt-4.1-nano"),
};

Recommended Models:

Model	Input / 1M tokens	Output / 1M tokens	Context
gpt-4.1-nano	$0.10	$0.40	1M
gpt-4.1-mini	$0.40	$1.60	1M
gpt-4.1	$2.00	$8.00	1M

3. Anthropic Claude

Get an API key from console.anthropic.com and set ANTHROPIC_API_KEY in your environment.

import { anthropic } from "@ai-sdk/anthropic";

export default {
  model: anthropic("claude-haiku-4-5-20250414"),
};

Recommended Models:

Model	Input / 1M tokens	Output / 1M tokens	Context
claude-haiku-4-5	$1.00	$5.00	200K
claude-sonnet-4-5	$3.00	$15.00	200K

4. Google Gemini

Set up a Google AI project and get an API key from aistudio.google.com. Set GOOGLE_GENERATIVE_AI_API_KEY in your environment.

import { google } from "@ai-sdk/google";

export default {
  model: google("gemini-2.5-flash"),
};

Recommended Models:

Model	Input / 1M tokens	Output / 1M tokens	Context
gemini-2.5-flash	$0.15	$0.60	1M
gemini-2.5-pro	$1.25	$10.00	1M

Best value for translations: gemini-2.5-flash — cheapest capable model with a 1M context window. gpt-4.1-nano is a close second if you're already in the OpenAI ecosystem.

Context Window Requirements

All models must support a minimum of 16,000 tokens context window. Every model listed above exceeds this by a wide margin.

If you encounter connection issues, verify your API key is set and your provider's server is accessible.