Token Intelligence

See where your team's
AI spend goes.

Forge reads recorded usage from Claude Code and supported Codex versions, with no proxy between your tools and the model. Compare estimated spend by model, workflow, and ticket. Estimates use published model rates; your actual subscription bill or negotiated pricing may differ.

Provider-neutral
One unit for every measured model and tool. Forge converts usage from every model into “weighted tokens,” so you can compare them directly, whether you pay a flat rate or per token.
Cross-tool
Compare Claude Code and Codex usage in one view. Cursor sessions are currently unmeasured because the tool does not expose per-session token usage to plugins.
By workflow and ticket
See spend by workflow and by ticket, plus cost per minute, which shows which work costs the most for the time it takes.
Example dashboard with sample data.
Token Intelligence Team
Sample
Covers sessions Forge recorded: workflow runs and, when tracking is on, work outside a workflow, with or without a ticket. Other AI activity and Cursor token usage aren’t measured here.
Estimated weighted tokens
167M
—
Estimated cost
$502
—

AI Spend Trend

167M total · 24M/day avg
MonTueWedThuFriSatSun

Token Composition

29.3× cache reuse
Fresh input
1.9M<1%
Cache reads
521M96%
Cache writes
18M3%
Output
2.9M<1%

By AI Tool & Model

Tool%WeightedCostSessions
Claude 96% 161M $483 12
Codex 4% 6.2M $18.64 7
Model%WeightedCostSessions
claude-opus-4-895%159M$47812
gpt-5.6-sol4%6.2M$18.587
claude-sonnet-4-6<1%1.0M$3.122
claude-haiku-4-5-20251001<1%573K$1.724
gpt-5.6-terra<1%14K$0.042
gpt-5.6-luna<1%6.0K$0.021

Tip: The No workflow row below shows AI work that ran outside a workflow. Compare its cost per minute with your workflows to see where a cheaper model or a set of steps could help.

By Workflow

Workflow%WeightedCostCost / min
Implement a Feature24%40M$119$0.86 
Shape a Feature13%22M$64.83$1.66 
Fix a Bug7%11M$34.34$0.65 
Update Work Item3%4.9M$14.59$0.63 
Explore Architecture1%1.9M$5.72$0.14 
Review a Pull Request<1%1.4M$4.24$0.28 
Check Work Item Status<1%1.2M$3.54$0.15 
Plan an Implementation<1%372K$1.12$0.22 
No workflow 51%85M$254$0.54 

Top tickets by AI spend

Ticket Title AI spend AI time Cost / min Tools
ENG-412 User onboarding flow redesign $12642M weighted 2h 41m $0.78 
Jun 11, 10:34 PM
$5.721.9M weighted 16m $0.35 
Jun 11, 11:36 PM
$2.50833K weighted 7m $0.34 
Jun 13, 11:10 PM
$10435M weighted 1h 14m $1.40 
Jun 14, 01:15 PM
$13.024.3M weighted 2m $6.37 
Fresh input 817K Cache reads 137M Cache writes 4.4M Output 660K Cache reuse31.3×

+1h of this ticket's time — spend booked under other tickets

ENG-401 Cross-tool token pricing $34.3411M weighted 53m $0.65 
ENG-388 Self-serve billing & plan management $15.045.0M weighted 55m $0.27 
ENG-339 API rate limiting & throttling $9.393.1M weighted 21m $0.43 
ENG-318 Audit log export to CSV $2.14712K weighted 6m $0.32 
ENG-356 Refactor notification service $1.40468K weighted 10m $0.13 

Forge estimates usage from recorded Claude Code and Codex sessions. Cursor usage is currently unmeasured. Estimates use published model rates and may differ from actual bills.

Cost control

Stop paying top-model prices
for routine work.

Every workflow step is marked for a fast, balanced, or capable model. Switching models has a cost, so Forge hands a step to a different model only when that pays off. Start your session on a balanced model, such as Sonnet in Claude Code, and use Token Intelligence to see where the money goes.

Model delegation is available in Claude Code and Codex. Token measurement covers Claude Code and supported Codex versions; Cursor usage is currently unmeasured.

Know what your AI costs
before the invoice arrives.

Token Intelligence is included with Team, free during beta. Install the plugin, then open Token Intelligence in the Forge dashboard.

Start free → See how to install