Skip to main content

Supported Models

VaultProxy provides access to models from 10+ AI providers through a single API endpoint. You specify the model in your request, and VaultProxy routes it to the correct provider automatically.

Model Naming Convention

Models can be referenced in two ways:

  • Bare name (for OpenAI models): gpt-4o, o1
  • Provider-prefixed: provider/model-name (e.g., anthropic/claude-4.6-sonnet, google/gemini-2.5-pro)

OpenAI models work with either format: gpt-4o and openai/gpt-4o are equivalent.

OpenAI

ModelContext WindowDescription
gpt-5.4256KMost capable OpenAI model
gpt-5.4-mini256KEfficient version of GPT-5.4
gpt-4o128KFast, multimodal flagship model
gpt-4o-mini128KAffordable small model for lightweight tasks
o1200KAdvanced reasoning model
o1-mini128KSmaller, faster reasoning model

Anthropic

ModelContext WindowDescription
anthropic/claude-4.6-opus1MMost powerful Claude model
anthropic/claude-4.6-sonnet1MBalanced performance and speed
anthropic/claude-4.6-haiku1MFastest Claude model

Google

ModelContext WindowDescription
google/gemini-3.1-pro2MLatest Gemini Pro model
google/gemini-2.5-pro1MPrevious generation Pro
google/gemini-2.5-flash1MFast and efficient
google/gemini-2.5-flash-lite1MLightest Gemini model

Mistral

ModelContext WindowDescription
mistral/mistral-large-3128KMistral flagship model
mistral/mistral-small-4128KEfficient small model
mistral/pixtral-large128KMultimodal with vision

Groq

ModelContext WindowDescription
groq/llama-4-scout128KLlama 4 on Groq inference
groq/llama-4-maverick128KLlama 4 Maverick variant
groq/deepseek-v364KDeepSeek V3 on Groq

DeepSeek

ModelContext WindowDescription
deepseek/deepseek-v3.2128KLatest DeepSeek general model
deepseek/deepseek-reasoner128KSpecialized reasoning model

Cohere

ModelContext WindowDescription
cohere/command-a256KLatest Command model
cohere/command-r-plus128KRAG-optimized model

xAI

ModelContext WindowDescription
xai/grok-4256KxAI flagship model

Perplexity

ModelContext WindowDescription
perplexity/pplx-70b-online128KSearch-augmented model with live data

Bielik

ModelContext WindowDescription
bielik/bielik-13b-pro8KPolish-focused LLM, 13B parameters
bielik/bielik-7b-v28KSmaller Polish-focused model
tip

Bielik models are specifically trained on Polish language data and may offer better performance for Polish-language tasks compared to general-purpose models.

Example: Using Different Models

# OpenAI (bare name)
curl -X POST https://api.vaultproxy.ai/v1/chat/completions \
-H "Authorization: Bearer vpx_live_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Hello"}]}'

# Anthropic (provider-prefixed)
curl -X POST https://api.vaultproxy.ai/v1/chat/completions \
-H "Authorization: Bearer vpx_live_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "anthropic/claude-4.6-sonnet", "messages": [{"role": "user", "content": "Hello"}]}'

# Bielik (provider-prefixed)
curl -X POST https://api.vaultproxy.ai/v1/chat/completions \
-H "Authorization: Bearer vpx_live_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "bielik/bielik-13b-pro", "messages": [{"role": "user", "content": "Cześć!"}]}'
info

You must have a valid provider key configured for the provider of the model you are requesting. If you request an Anthropic model but have not added an Anthropic provider key, the request will return a 400 error.