Native Tools Calling

This guide provides an overview of which providers and models natively support tool calling capabilities. The information is organized by provider to help you choose the right model for your needs.

Models with Native Support

The MakeHub API includes information about native tool calling capabilities in the model metadata. You can use the /v1/models endpoint to identify which models support these features natively.

Filtering Models by Capability

You can use the model information to filter models based on specific capabilities:

def find_models_with_tools_calling():
    models_data = list_models()
    
    tools_calling_models = [
        model for model in models_data["data"] 
        if model.get("native_support_tools_calling", False)
    ]
    
    return tools_calling_models
 
# Find models that support both tool calling and streaming
def find_streaming_tool_calling_models():
    models_data = list_models()
    
    filtered_models = [
        model for model in models_data["data"] 
        if model.get("native_support_tools_calling", False) and 
           model.get("streaming_tool_calling_support", False)
    ]
    
    return filtered_models

Providers and Their Tool Calling Support

Below is a comprehensive overview of providers and their models, showing which ones support tool calling capabilities and which don't.

Anthropic

Model	Tool Calling	Streaming Tool Calling
claude-3-5-haiku	✅	✅
claude-3-5-sonnet	✅	✅
claude-3-7-sonnet	✅	✅

Azure

Model	Tool Calling	Streaming Tool Calling
gpt-4o	✅	✅
gpt-4o-mini*	✅	✅

*Available in UKSouth, norwayeast, swedencentral, switzerlandnorth regions

Bedrock

Model	Tool Calling	Streaming Tool Calling
claude-3-5-haiku	✅	✅
claude-3-5-sonnet	✅	✅
claude-3-7-sonnet	✅	✅

CentML

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-fp8	❌	❌
Llama-3.3-70B-fp16	✅	✅

Cerebras

Model	Tool Calling	Streaming Tool Calling
Llama-3.1-8B-fp16	✅	✅
Llama-3.3-70B-fp16	✅	✅

ChutesAI

Model	Tool Calling	Streaming Tool Calling
deepseek-V3-0324-fp8	✅	❌

DeepInfra

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-distill-llama-70b-fp16	❌	❌
deepseek-R1-fp8	✅	✅
deepseek-V3-0324-fp8	✅	❌
deepseek-V3-fp8	✅	❌
gemma-2-9b-fp16	❌	❌
gemma-3-27B	❌	❌
Llama-3.1-70B-fp16	✅	✅
Llama-3.1-8B-fp16	✅	✅
Llama-3.1-8B-fp8	✅	✅
Llama-3.3-70B-fp16	✅	✅
Llama-3.3-70B-fp8	✅	✅
mistral-small-24B-fp8	❌	❌
open-mistral-nemo	❌	❌
QWQ-32B-fp8	❌	❌
Qwen2.5-Coder-32B	❌	❌

Deepseek

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-fp8	❌	❌
deepseek-V3-fp8	✅	✅

Fireworks

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-fp8	❌	❌
deepseek-V3-0324-fp8	✅	✅
deepseek-V3-fp8	✅	✅
Llama-3.1-70B-fp16	✅	✅
Llama-3.1-8B-fp16	❌	❌
Llama-3.3-70B-fp16	❌	❌
mistral-small-24B-fp16	❌	❌
QWQ-32b-fp16	❌	❌
Qwen2.5-Coder-32B	❌	❌

Google

Model	Tool Calling	Streaming Tool Calling
gemini-2.0-flash	❌	❌
gemini-2.0-flash-lite-preview	✅	✅
gemini-2.0-flash-thinking	❌	❌
gemini-2.5-pro-exp-03-25	✅	✅

Groq

Model	Tool Calling	Streaming Tool Calling
Llama-3.3-70B-fp16	✅	✅

Hyperbolic

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-fp8	✅	✅
deepseek-V3-0324-fp8	✅	✅
deepseek-V3-fp8	✅	✅
Llama-3.1-70B-fp16	✅	✅
Llama-3.1-8B-fp16	✅	✅
Llama-3.3-70B-fp16	✅	✅
QWQ-32b-fp16	❌	❌

Leptonai

Model	Tool Calling	Streaming Tool Calling
Llama-3.1-70B-fp8	✅	❌
Llama-3.1-8B-fp8	✅	❌
Llama-3.3-70B-fp8	✅	❌

Mistral

Model	Tool Calling	Streaming Tool Calling
mistral-small-24B-fp16	✅	❌
open-mistral-nemo	✅	❌

Nebius-base

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-fp8	❌	❌
deepseek-V3-0324-fp8	❌	❌
deepseek-V3-fp8	❌	❌
Llama-3.1-70B-fp16	✅	✅
Llama-3.1-8B-fp16	✅	✅
open-mistral-nemo	❌	❌
QWQ-32b-fp16	❌	❌

Nebius-fast

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-fp8	❌	❌
deepseek-V3-0324-fp8	❌	❌
Llama-3.1-70B-fp16	✅	✅
Llama-3.1-8B-fp16	✅	✅
open-mistral-nemo	❌	❌
QWQ-32b-fp16	❌	❌

Novitai

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-fp8	✅	✅
deepseek-V3-fp8	✅	✅
gemma-2-9b-fp16	❌	❌
Llama-3.1-70B-fp16	❌	❌
Llama-3.1-8B-fp16	❌	❌
Llama-3.3-70B-fp16	✅	✅
mistral-7B-v0.3	❌	❌
open-mistral-nemo	❌	❌

OpenAI

Model	Tool Calling	Streaming Tool Calling
gpt-3.5-turbo	✅	✅
gpt-4.5	✅	✅
gpt-4o	✅	✅
gpt-4o-mini	✅	✅

Replicate

Model	Tool Calling	Streaming Tool Calling
claude-3-5-sonnet	✅	✅

Sambanova

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-distill-llama-70b-fp16	❌	❌
deepseek-R1-fp8	❌	❌
deepseek-V3-0324-fp8	❌	❌
Llama-3.1-8B-fp16	✅	✅
Llama-3.3-70B-fp16	✅	✅

Together

Model	Tool Calling	Streaming Tool Calling
deepseek-R1-fp8	✅	❌
deepseek-V3-fp8	✅	❌
MythoMax-L2-13b-fp16	✅	❌
Llama-3.1-70B-fp8	✅	✅
Llama-3.1-8B-fp8	✅	✅
Llama-3.3-70B-fp8	✅	✅
mistral-small-24B-fp16	✅	❌
QWQ-32b-fp16	✅	✅
Qwen2.5-Coder-32B	✅	✅

XAI

Model	Tool Calling	Streaming Tool Calling
grok-2	✅	✅
grok-beta	✅	✅

Benefits of Native Support

Models with native support for tool calling typically offer:

Better parsing of tool arguments
More reliable adherence to tool schemas
Optimized performance for tool-use scenarios
Reduced likelihood of hallucinated parameters

When building applications that require tool calling, choosing models with native support can lead to more robust and predictable behavior.

Usage Considerations

When using tool calling capabilities:

Regular tool calling is suitable for most applications where immediate response streaming isn't required
Streaming tool calling is ideal for interactive applications where real-time responses enhance user experience
Consider your specific use case requirements when choosing between regular and streaming implementations
Models marked with ❌ either don't support tool calling or failed during testing

Family Models MakeHub Tools Layer