ClearBurn: LLM Token & Cost Estimator for VS Code

See the token count and $ cost of any prompt, file, or selection before you call the API. No more finding out you burned $40 on Opus after the fact.

Why this exists

If you build with Claude, GPT, or Gemini APIs, you already know the pain. Pricing changes almost every month, across four or more vendors, each with different input/output rates. You don't know the real cost of a prompt until you've already sent it. Comparing which model is cheapest for a given job usually means opening three pricing pages and doing mental math.

ClearBurn puts the answer inline, in the editor, in one click.

Free

Live inline cost (CodeLens). The moment you open a file, every prompt-like string (system prompts, template literals used as messages, etc.) gets a small "⚡ ~1,240 tok · ~$0.018/call" line right above it. No selecting text, no running a command. It's just there, and it updates as you type. This is the core feature and it needs zero learning curve: if you can read a CodeLens (same UI as "2 references" above a function), you already know how to use this.

Estimate Cost of Selection / Current File. Manual commands for one-off checks outside a detected prompt, for example checking a chunk of a big document before pasting it into a prompt.

Model auto-detection. If your code has model: "claude-sonnet-5" (or similar) near the prompt, the CodeLens prices it against that exact model automatically instead of a generic default.

Pro ($9 one-time, license key)

Scan Workspace for AI Cost Exposure. One command scans your whole repo (.ts, .tsx, .js, .jsx, .py, skips node_modules, dist, etc.), finds every prompt call site, and shows a sorted report: which prompt costs the most per call, plus a projected monthly cost based on a call volume you control (clearBurn.assumedMonthlyCallsPerSite). This answers the question an individual developer's mental math can't: where is our AI spend actually going, once a project has fifteen or more call sites across a codebase.

Compare Cost Across All Models. Click any inline lens to see the same prompt priced across every model side by side.

Project Running Total. A status bar counter tracking cumulative estimated spend across your session, with full history.

Export CSV. Dump the cost log for expense reports or budgeting.

Pricing table stays current through the clearBurn.customPricing setting, editable anytime, no extension update needed.

Why this is more than a calculator

The free tier answers "what does this one prompt cost." Useful, but something a developer can half-estimate mentally on its own. The Pro tier answers two questions nobody else answers well before the bill arrives: across my whole codebase, where is AI spend concentrated, and what will this cost me at real production volume, not just once.

That reframes the product from an individual utility into something closer to a lightweight FinOps and cost-governance tool for teams shipping LLM features.

How Pro unlocking works

Pro features unlock with a license key, delivered by email right after purchase. Paste it into the "ClearBurn: Enter License Key" command and Pro features activate immediately, no restart needed. It also works offline between checks, so it won't lock you out mid-flight.

Privacy and network use

ClearBurn makes exactly one network call: to Gumroad's license-verification API, and only when you enter or refresh a Pro license key. Token counting, pricing, and the workspace scan all run locally. Your prompts and code never leave your machine. No telemetry, no analytics, no tracking.

Support

Built and maintained by Artcllick.ro (www.artcllick.ro). Questions, bugs, or feature requests: artclickon@outlook.com. Read personally, not a support queue.

Token counting accuracy note

Token counts use a BPE tokenizer (gpt-tokenizer) as a cross-vendor approximation. It's exact for OpenAI-style tokenization and a reasonable estimate, typically within 10 to 20 percent, for Claude and Gemini, since Anthropic and Google don't publish an offline tokenizer. This is stated here rather than hidden, because accuracy honesty is a retention feature for a paid tool, not a weakness to bury.