Model Providers

Tinbox supports four major AI model providers, each with unique strengths. Choose your provider based on quality requirements, cost constraints, and infrastructure preferences.

Supported Providers

OpenAI

GPT-4o and GPT-5 models with vision and extended thinking support

Anthropic

Claude models with strong multilingual capabilities

Google

Gemini models with competitive pricing and performance

Ollama

Local models with zero API costs and full privacy

Model Types

From types.py:21-28:

class ModelType(str, Enum):
    """Supported LLM model providers."""
    
    OPENAI = "openai"      # OpenAI models (GPT-4, etc.)
    ANTHROPIC = "anthropic"  # Anthropic models (Claude)
    OLLAMA = "ollama"      # Local models via Ollama
    GEMINI = "gemini"      # Google's Gemini models

OpenAI

OpenAI provides state-of-the-art language models with strong multilingual support and vision capabilities.

Available Models

GPT-5 (2025-08-07): Latest flagship model with extended thinking
GPT-4o: Fast and cost-effective with vision support
GPT-4-turbo: Previous generation turbo model

Pricing

From cost.py:22-27:

ModelType.OPENAI: (
    0.00125,  # $0.00125 per 1K input tokens
    0.01,     # $0.01 per 1K output tokens (GPT-5)
)

CLI Usage

# Default GPT-5 model
tinbox translate --to de --model openai document.pdf

# Specific model version
tinbox translate --to de --model openai:gpt-4o document.pdf

# With extended thinking (reasoning models)
tinbox translate --to de --model openai:gpt-5-2025-08-07 \
  --reasoning-effort high \
  document.pdf

Environment Setup

export OPENAI_API_KEY="sk-..."

GPT-5 with extended thinking (--reasoning-effort high) produces exceptional translation quality but increases cost and time significantly. Use for critical documents only.

Features

✅ Vision support (for PDF images)
✅ Extended thinking (reasoning models)
✅ Checkpoint/resume
✅ Glossary support
✅ Fast inference
✅ High quality output

Anthropic

Anthropic’s Claude models excel at nuanced language understanding and long-form content translation.

Available Models

Claude Sonnet 4: Latest model with extended thinking
Claude 3.5 Sonnet: Fast and cost-effective
Claude 3 Opus: Highest quality (previous generation)

Pricing

From cost.py:28-31:

ModelType.ANTHROPIC: (
    0.003,    # $0.003 per 1K input tokens
    0.015,    # $0.015 per 1K output tokens (Sonnet 4)
)

CLI Usage

# Default Claude Sonnet 4
tinbox translate --to de --model anthropic document.pdf

# Specific model version
tinbox translate --to de --model anthropic:claude-3-5-sonnet document.pdf

# With extended thinking
tinbox translate --to de --model anthropic:claude-sonnet-4-20250514 \
  --reasoning-effort medium \
  document.pdf

Environment Setup

export ANTHROPIC_API_KEY="sk-ant-..."

Claude models are particularly strong at maintaining consistent tone and style across long documents. Ideal for literary translations.

Features

✅ Vision support (for PDF images)
✅ Extended thinking (reasoning models)
✅ Checkpoint/resume
✅ Glossary support
✅ Strong multilingual support
✅ Excellent at literary translation

Google (Gemini)

Google’s Gemini models offer competitive performance with cost-effective pricing.

Available Models

Gemini 2.5 Pro: Latest model with strong multilingual support
Gemini 1.5 Pro: Previous generation
Gemini Flash: Fast and economical

Pricing

From cost.py:32-35:

ModelType.GEMINI: (
    0.00125,  # $0.00125 per 1K input tokens
    0.01,     # $0.01 per 1K output tokens (Gemini 2.5 Pro)
)

CLI Usage

# Default Gemini 2.5 Pro
tinbox translate --to de --model gemini document.pdf

# Specific model version
tinbox translate --to de --model gemini:gemini-2.5-flash document.pdf

Environment Setup

export GOOGLE_API_KEY="AIza..."

Gemini models offer the same pricing as OpenAI but with different strengths. Test both to see which works better for your use case.

Features

✅ Vision support (for PDF images)
✅ Checkpoint/resume
✅ Glossary support
✅ Fast inference
✅ Cost-effective pricing

Ollama (Local Models)

Ollama runs models locally on your machine, providing zero-cost translation with complete privacy.

Available Models

Any model available in Ollama can be used:

Llama 3: Strong general-purpose model
Mistral: Fast and efficient
Gemma: Google’s open model
Qwen: Excellent multilingual support

Pricing

From cost.py:36:

ModelType.OLLAMA: (0.0, 0.0),  # Free for local models

Setup

Install Ollama: https://ollama.ai/download
Pull a model:

ollama pull llama3:8b

Translate:

tinbox translate --to de --model ollama:llama3:8b document.txt

CLI Usage

# Using Llama 3
tinbox translate --to de --model ollama:llama3:8b document.txt

# Using Qwen for multilingual
tinbox translate --to ja --model ollama:qwen2:7b document.txt

# Using Mistral
tinbox translate --to es --model ollama:mistral:7b document.txt

Local models don’t support vision, so PDF files must be processed as text. They also don’t support extended thinking (reasoning).

Features

✅ Zero cost
✅ Complete privacy
✅ Checkpoint/resume
✅ Glossary support
✅ Offline operation
❌ No vision support
❌ No extended thinking
⚠️ Slower than cloud models
⚠️ Quality varies by model

Performance Considerations

From cost.py:198-200:

# Estimate time (very rough estimate)
# Assume 30 tokens/second for cloud models, 20 tokens/second for local
tokens_per_second = 20 if model == ModelType.OLLAMA else 30

Use Ollama for large documents where cost would otherwise be prohibitive. The quality is acceptable for most use cases, especially with larger models (13B+).

Model Selection Syntax

Tinbox supports flexible model selection:

# Provider only (uses default model)
tinbox translate --to de --model openai document.pdf

# Provider with specific model
tinbox translate --to de --model openai:gpt-4o document.pdf

# Provider with model and version
tinbox translate --to de --model openai:gpt-5-2025-08-07 document.pdf

# Ollama with model and size
tinbox translate --to de --model ollama:llama3:8b document.txt

Reasoning Effort Levels

Extended thinking models (OpenAI GPT-5, Anthropic Sonnet 4) support reasoning effort configuration: From types.py:110-113:

reasoning_effort: Literal["minimal", "low", "medium", "high"] = Field(
    default="minimal",
    description="Model reasoning effort level. Higher levels improve quality but increase cost and time significantly.",
)

Effort Levels

minimal (default): Standard translation, no extended thinking
low: Brief reasoning, 2-3x cost
medium: Moderate reasoning, 5-10x cost
high: Deep reasoning, 10-20x cost

# Minimal (default)
tinbox translate --to de --model openai document.pdf

# Low reasoning
tinbox translate --to de --model openai --reasoning-effort low document.pdf

# High reasoning for critical translation
tinbox translate --to de --model openai --reasoning-effort high \
  --max-cost 50.00 \
  document.pdf

Higher reasoning effort dramatically increases both cost and time. Always set --max-cost when using medium or high effort levels.

Cost Tracking

All models (except Ollama) track real-time costs during translation. See Cost Tracking for details.

Choosing the Right Provider

When to use OpenAI

You need vision support for PDF translation
You want the highest quality output
You need extended thinking for critical translations
Cost is not the primary concern

When to use Anthropic

You’re translating literary or creative content
You need consistent tone and style
You value nuanced language understanding
You want extended thinking capabilities

When to use Google (Gemini)

You want competitive quality at lower cost
You need fast inference
You’re translating general-purpose content
You want an alternative to OpenAI

When to use Ollama

Cost is a critical factor
You need complete privacy
You’re translating large volumes
You have local compute resources
You’re willing to accept lower quality

Provider Comparison

Feature	OpenAI	Anthropic	Google	Ollama
Cost (1K tokens)	$0.00125-$ 0.01	$0.003-$ 0.015	$0.00125-$ 0.01	Free
Vision	✅ Yes	✅ Yes	✅ Yes	❌ No
Reasoning	✅ Yes	✅ Yes	❌ No	❌ No
Speed	Fast	Fast	Fast	Slow
Quality	Excellent	Excellent	Very Good	Good
Privacy	Cloud	Cloud	Cloud	Local
Offline	❌ No	❌ No	❌ No	✅ Yes

Get Started

Core Concepts

Guides

Advanced

​Supported Providers

OpenAI

Anthropic

Google

Ollama

​Model Types

​OpenAI

​Available Models

​Pricing

​CLI Usage

​Environment Setup

​Features

​Anthropic

​Available Models

​Pricing

​CLI Usage

​Environment Setup

​Features

​Google (Gemini)

​Available Models

​Pricing

​CLI Usage

​Environment Setup

​Features

​Ollama (Local Models)

​Available Models

​Pricing

​Setup

​CLI Usage

​Features

​Performance Considerations

​Model Selection Syntax

​Reasoning Effort Levels

​Effort Levels

​Cost Tracking

​Choosing the Right Provider

​Provider Comparison

Build docs developers (and LLMs) love

Supported Providers

Model Types

OpenAI

Available Models

Pricing

CLI Usage

Environment Setup

Features

Anthropic

Available Models

Pricing

CLI Usage

Environment Setup

Features

Google (Gemini)

Available Models

Pricing

CLI Usage

Environment Setup

Features

Ollama (Local Models)

Available Models

Pricing

Setup

CLI Usage

Features

Performance Considerations

Model Selection Syntax

Reasoning Effort Levels

Effort Levels

Cost Tracking

Choosing the Right Provider

Provider Comparison