toolcalling.ai
Open workspace
function calling · tool use · arena

Find the model that calls your tools right.

A workspace for building and debugging LLM function calls, plus a community leaderboard ranked by blind votes on real tool-calling tasks.

Start for free View leaderboard
arena.json
1// battle #4821 · get_weather · votes: A2{3  "model_a": "███████████",4  "tool_calls": [{5    "name": "get_weather",6    "arguments": { "city": "Istanbul",7                   "unit": "celsius" }8  }],9  "schema_valid": true,10  "latency_ms": 81211}
Developers
1
Model runs
0
Arena votes
0
Models ranked
0

Top models for function calling

Arena score from blind pairwise votes

Full leaderboard
Rankings appear after the first arena votes.

Top raters

Most votes cast by the community

Be the first to vote.

How it works

Build in the playground with your own keys. When a result surprises you, compare it against a random model on ours. Every vote improves the public ranking.

  1. 1 · Build
    Define tools and a conversation, then run it on any model with your API key.
  2. 2 · Rate
    Thumbs up or down, or compare it against a random model blind (a few free battles a day).
  3. 3 · Earn
    Votes rank models and providers. Streaks and invites earn you points.

Learn function calling

A six-step recipe from our docs at toolaugmented.ai: design schemas, write system instructions, shape tool outputs, and measure.

All guides
  1. 1Understand the loop
    What is function calling
  2. 2Design your tools
    Schema best practices
  3. 3Write the system instruction
    System instruction
  4. 4Shape tool outputs
    Tool output
  5. 5Cover the call patterns
    Function call patterns
  6. 6Pick a model and measure
    Evaluation

Tool-calling editor

System prompt, JSON-Schema tools, and user / assistant / tool-call / tool-result turns you can edit and re-run.

Every major provider

OpenAI, Anthropic, Mistral, Groq, DeepSeek, Together, OpenRouter, and local Ollama / llama.cpp / vLLM.

Blind model arena

Same conversation, two anonymous models. Vote, then see who was who.

Schema-checked calls

Every tool call is validated against its schema, so rankings reflect valid calls, not just good-sounding text.

Full history

Every conversation and run is saved. Search, fork, share a link, or export to JSON and notebooks.

Points & streaks

Earn points for 5-day streaks and invited friends, and spend them on advanced features.

Ready to test your tools?

Free account. Your API keys never leave the request.

Open the workspace