tokentoll LLM Cost Diff

May 4, 2026

Version updated for https://github.com/Jwrede/tokentoll to version v0.6.1.

This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

tokentoll is a CLI tool and GitHub Action designed to analyze code for Large Language Model (LLM) API calls, estimate their associated costs, and highlight cost impacts of code changes directly in pull requests or the terminal. It helps developers identify potential cost implications, such as increased API expenses due to model changes or new calls, during the code review process. Key features include static analysis of various LLM SDKs, cost estimation, and detailed cost comparisons between commits or branches.

What’s Changed

Summary

Reverts the v0.6.0 special-case that skipped AzureChatOpenAI(deployment_name=...) calls. The skip was inconsistent: every other dynamic model gets a per-SDK default with a (default) label, only Azure was being silently dropped. v0.6.1 restores symmetry and adds an opt-in for users who want stricter behavior.

Changes

New: `skip_dynamic_models` config option

Set in .tokentoll.yml to suppress cost estimates for any call whose model name cannot be resolved statically:

skip_dynamic_models: true

Also supported as a per-path override:

overrides:
  - path: src/azure/
    skip_dynamic_models: true

When enabled, dynamic-model calls are still detected and listed but reported with no cost rather than priced against a default.

LangChain detector

AzureChatOpenAI(deployment_name=...) and AzureChatOpenAI(azure_deployment=...) calls without an explicit model= argument now flow through the standard dynamic-default path (priced as gpt-4o (default) unless overridden). Use the new skip_dynamic_models option to suppress these.

Upgrade

- uses: Jwrede/tokentoll@v0.6.1

pip install --upgrade tokentoll