models

Call counts, token usage, and cost broken down per model.

Usage

claudex models [-p/--project <substr>] [--json]

Flags

Flag	Description
-p, --project <substr>	Filter by substring match on the project path.
--json	Emit JSON.

Example

# Global breakdown
claudex models

# One project
claudex models --project claudex

# All Opus rows (any variant)
claudex models --json | jq '.[] | select(.model_family == "Opus")'

# Top model by cost
claudex models --json | jq 'sort_by(-.cost_usd)[0] | {model, cost_usd}'

Columns

Column	Source
Model	Full model tag from Claude Code (e.g. claude-opus-4-7, claude-sonnet-4-6).
Family	Opus / Sonnet / Haiku (derived from the name).
Sessions	Distinct sessions that used the model.
Input	Total input tokens.
Output	Total output tokens.
Cache Write	Total cache-creation tokens.
Cache Read	Total cache-read tokens.
Avg/Session	Average spend per session using the model.
Avg Tokens	Average total tokens per session using the model.
Cost (USD)	Model-specific cost.

JSON shape

[
  {
    "model": "claude-opus-4-6",
    "model_family": "Opus",
    "session_count": 467,
    "input_tokens": 1050304,
    "output_tokens": 21225014,
    "cache_creation_tokens": 4027298,
    "cache_read_tokens": 893758965,
    "avg_cost_per_session_usd": 64.35,
    "avg_tokens_per_session": 1958324.4,
    "service_tiers": ["default", "priority"],
    "inference_geos": ["us-east-1", "eu-west-1"],
    "avg_speed": 35.2,
    "total_iterations": 982,
    "cost_usd": 30051.131094
  }
]

Notes

• Family detection. model_family is a substring match on the model tag: anything containing opus is Opus, haiku is Haiku, anything else is Sonnet. See Pricing model.
• Mixed-model sessions. A session that switched models appears under every model it used; session_count counts each model-session pair once.
• Runtime metadata. service_tiers, inference_geos, avg_speed, and total_iterations come from the usage blocks Claude Code records on assistant messages. avg_speed is the mean of per-session model averages, not a throughput-weighted global average. These fields are best-effort; missing values stay empty/null.