LLM API Cost Analyzer: Understand Your AI Usage Bill
Upload API usage data to see cost by model, day, and project, cache savings, spikes, and month forecast with custom prices. 100% private in your browser.
Model Price Table (per 1M Tokens USD)
| Model Pattern (Prefix*) | Prompt $/1M | Completion $/1M | CacheRead $/1M | Action |
|---|
| Model | Calls | Prompt Tokens | Completion Tokens | Total Cost | Spend Share |
|---|---|---|---|---|---|
| No logs analyzed yet | |||||
| Date (UTC) | Calls | Total Tokens | Daily Cost | Cache Saved |
|---|---|---|---|---|
| No logs analyzed yet | ||||
| Masked Key / Project | Calls | Total Tokens | Total Cost | Share |
|---|---|---|---|---|
| No logs analyzed yet | ||||
| Timestamp (UTC) | Model | Prompt Tokens | Completion Tokens | Cost (USD) | Flag |
|---|---|---|---|---|---|
| No logs analyzed yet | |||||
How to Audit and Reduce LLM API Spend
Large Language Model architectures bill on asymmetric multi-tier token pricing. Between differing rates for prompt input, cached prefixes, visible output, and hidden reasoning traces, raw monthly billing statements rarely make it obvious where money is going:
Prompt Caching Economics
Modern models (Claude 3.5 Sonnet, GPT-4o, DeepSeek) discount repeated prompt tokens by 80% to 90%. Decomposing cached reads highlights which endpoints leverage system prompt caching effectively.
3-Sigma Spike Outliers
Recursive agent tool calls or unbounded conversation contexts can silently trigger $5+ single requests. Our statistical 3-sigma anomaly detector flags these spikes automatically.
Private & Offline
No tokens, trace logs, or API keys ever touch remote servers. Parsing, currency conversions, and rate matrix evaluations execute 100% inside your local browser tab.
Frequently Asked Questions
Which file formats can I upload?
The analyzer automatically ingests CSV exports from OpenAI or Anthropic consoles, line-delimited JSONL trace files, and standard JSON arrays.
How are API keys and project names secured?
All sensitive keys matching API patterns (like sk-proj-... or sk-ant-...) are immediately masked client-side to only display their final 4 characters.
Can I customize token price tiers?
Yes. Open the Rates & Schema drawer to add custom model wildcards, override prompt and completion costs per 1M tokens, or import custom JSON rate schemes.