vscode-extension

Claude Cost Estimator - VS Code Extension

A lightweight VS Code extension that shows estimated token count and Claude API cost for the active file, directly in the status bar. All estimation runs locally with no API calls.

Features

• Status bar token count: Shows ~1,247 tokens | $0.004 for the active file, updated in real time as you type.
• Estimate Current File: Detailed breakdown of input/output costs across all Claude model tiers (Fable 5, Opus 5, legacy Opus snapshots, Sonnet 5, Haiku 4.5).
• Estimate CLAUDE.md Per-Turn Cost: Finds all CLAUDE.md files in the workspace and calculates how much they add to each turn of a Claude Code session. Warns when the file is too small to be cacheable on the selected model.
• Check Project Configuration: Audits your workspace for CLAUDE.md, .claudeignore, and .claude/settings.json. Reports missing files and suggests improvements.
• CLAUDE.md size warning: The status bar turns yellow when a CLAUDE.md file exceeds the configured line threshold, since large instruction files increase per-turn costs.

Installation

This extension is not yet published to the VS Code Marketplace. To run it in development mode:

cd tools/vscode-extension
npm install
npm run compile

Then open VS Code, press F5 (or Run > Start Debugging), and select "VS Code Extension Development Host". The extension will activate in the new window.

Alternatively, to install locally as a VSIX:

npm install -g @vscode/vsce
cd tools/vscode-extension
npm install
npm run compile
vsce package
code --install-extension claude-cost-estimator-0.1.0.vsix

Commands

Open the command palette (Ctrl+Shift+P / Cmd+Shift+P) and search for:

Command	Description
Claude Cost: Estimate Current File	Show token count and cost breakdown for the active file
Claude Cost: Estimate CLAUDE.md Per-Turn Cost	Calculate per-turn and 20-turn session costs for all CLAUDE.md files
Claude Cost: Check Project Configuration	Audit workspace for Claude Code config files

Configuration

Setting	Default	Description
claudeCost.defaultModel	sonnet	Model for cost estimation: fable, opus (Opus 5), opus-4.8, opus-4.7, opus-4.6, sonnet (Sonnet 5), or haiku
claudeCost.showInStatusBar	true	Show the token count status bar item
claudeCost.claudeMdWarningThreshold	150	Line count above which CLAUDE.md triggers a warning

How Token Estimation Works

The extension uses a character-based heuristic rather than a real BPE tokenizer:

• Code files average ~4 characters per token
• Prose files average ~3.5 characters per token
• The ratio is blended based on how many syntax characters the file contains

This is fast and dependency-free, with roughly 10-15% variance compared to the actual tokenizer. For precise counts, use the tools/token-estimator CLI which uses tiktoken.

Pricing Data

Costs are based on Claude API pricing as of 2026-07-25:

Model	Input (per 1M tokens)	Output (per 1M tokens)	Min cacheable prompt
Fable 5 (alias: fable)	$10.00	$50.00	512
Opus 5 (alias: opus)	$5.00	$25.00	512
Opus 4.8, legacy (alias: opus-4.8)	$5.00	$25.00	1,024
Opus 4.7, legacy (alias: opus-4.7)	$5.00	$25.00	2,048
Opus 4.6, legacy (alias: opus-4.6)	$5.00	$25.00	4,096
Sonnet 5 (alias: sonnet)	$3.00	$15.00	1,024
Haiku 4.5 (alias: haiku)	$1.00	$5.00	4,096

The opus alias points at Opus 5 (GA 2026-07-24), which costs exactly what Opus 4.8 did. Opus 4.8, 4.7, and 4.6 stay in the table as legacy snapshots so you can price an older pinned model, not because you should pick one.

Sonnet 5 also has an introductory rate of $2/$10 through 2026-08-31. The table uses the standard $3/$15 so projections stay valid past that date.

Minimum cacheable prompt. A cache_control block on a prefix shorter than the listed floor is silently ignored: no error, no cache_creation_input_tokens, and full input price on every turn. "Estimate CLAUDE.md Per-Turn Cost" adds a note when your CLAUDE.md falls below the floor for the selected model. Watch this on Haiku 4.5 especially -- its 4,096-token floor is 8x Opus 5's, so a CLAUDE.md that caches fine on Opus quietly pays full price on Haiku.

Opus 5 note. Opus 5 has adaptive thinking on by default and reasoning tokens bill as output. The per-turn output figure here assumes a flat 500 tokens, which is a floor rather than an average on Opus 5. Lower output_config.effort for routine turns, or set thinking to disabled (legal only at effort high or below).

Publishing to the Marketplace (Future)

To publish as a proper VS Code extension:

1. Create a publisher account at https://marketplace.visualstudio.com/manage
2. Add a publisher field to package.json
3. Add an icon (128x128 PNG) and set the icon field
4. Run vsce publish

See the VS Code publishing docs for details.