Transparent Token Initiative

The fact that tokenlens is successfull validates that developers care about token usage and context management.

Over 100k developers rely on tokenlens, yet tokenlens can not ensure the data is correct.

We built on top of open sources like OpenRouter and models.dev. If their data is not correct or incomplete, tokenlens is not correct or incomplete.

What is this initiative about?

There are two main cases, where developers rely on fresh model data.

• cost tracking (for example via PostHog and Langfuse)
• context management

These information should be provided by the model providers themselves, not by third parties.

Milestones

• agree on a common format for the model data
• cooperate with model providers to implement the common format

Proposal

Previous formats are lacking information.

{
  // ... other fields ...
  "usage": {
    "input_tokens": 32,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens": 18,
    "output_tokens_details": {
      "reasoning_tokens": 0
    },
    "total_tokens": 50
  }
  // ... other fields ...
}

We propose to use the following format for the model data:

{
  // ... other fields ...
  "usage": {
    "input_tokens": 32,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens": 18,
    "output_tokens_details": {
      "reasoning_tokens": 0
    },
    "total_tokens": 50
  },
  "limit": {
    "context": 128000,
    "input": 120000,
    "output": 8000
  },
  "usage_costs": {
    "input_usd": 0.000048,
    "output_usd": 0.000108,
    "reasoning_usd": 0,
    "cache_read_usd": 0,
    "cache_write_usd": 0,
    "total_usd": 0.000156,
    "rates_used": {
      "input_per_mtokens_usd": 150,
      "output_per_mtokens_usd": 600,
      "reasoning_per_mtokens_usd": 0,
      "cache_read_per_mtokens_usd": 75,
      "cache_write_per_mtokens_usd": 300,
      "currency": "USD"
    }
  }
  // ... other fields ...
}

Signed by

Nicklas Scharpff (creator of tokenlens)

How can you help?

Share this initiative with your friends and colleagues.

If you are a model provider, please implement the new format.

If you are a developer, please share this initiative with your friends and colleagues.