Nanotokens
AI Token Economics Platform
Your AI costs more than its token count suggests.
Token reduction and cost reduction have a Pearson correlation of just r = 0.15. Nanotokens tracks real billed spend — across cache tiers, model versions, and platforms — so your team optimises what actually matters.
Research findings that shaped this product
Pearson r between token savings and cost savings
Weinberger & Hozez, 2026
Billed cost rise despite 38% fewer tool-output tokens
arXiv 2607.12161
Inference cost attributable to prompt-cache operations
Anthropic Pricing Docs
The compression trap. Weinberger & Hozez (arXiv 2607.12161) showed that aggressive tool-output compression cut delivered tokens by 38.4% but raised billed cost by 6.8%. Prompt-cache creation and reads dominate the input-side bill; compressing tool output barely touches them.
Trajectory inflation. Compression changes agent behaviour — adding retrieval, diagnosis, and test turns that offset local savings. LLMLingua (Jiang et al., 2023) achieves 20× prompt compression, but the downstream trajectory cost must be measured, not assumed away.
The right metric. Cost-per-successful-task (CPST) captures all of this: the billed dollars divided by the number of jobs completed correctly. Nanotokens surfaces CPST alongside raw token counts so you can tell whether an optimisation actually helped.
Everything your team needs to govern AI spend
As organisations scale AI across distributed teams, token costs compound non-linearly. Nanotokens gives every team member a private, accurate view of what they are spending and why.
Real billed cost, not token counts
Every log entry auto-computes the dollar amount using current provider pricing — including cache-write and cache-read tiers that raw token counts ignore.
Cache efficiency breakdown
Cache operations are the single largest driver of inference cost at scale. See hit rates, write costs, and read savings side by side.
Cost-per-successful-task (CPST)
Grounded in Weinberger & Hozez (2026): measure what matters — dollars per completed job, not tokens consumed.
Cross-platform comparison
Track Claude, OpenAI, Gemini, Codex, Antigravity, and any future provider on a single private dashboard. Prices update when you do.
Compression ROI analysis
LLMLingua and similar tools compress prompts — but longer agent trajectories can negate the savings. Nanotokens surfaces that tradeoff.
Team and enterprise tracking
Enterprise tier adds an API ingestion endpoint and spend-threshold webhooks so cost governance runs automatically as AI scales.
Supported platforms at launch
Add any future platform with custom pricing — the schema never locks you in.
Simple, transparent pricing
We track AI costs — of course we keep our own pricing clear.
Analyst
- ✓Unlimited platforms
- ✓Full history
- ✓Cache & compression analytics
- ✓CPST metric
- ✓CSV export
- ✓Custom model pricing
Enterprise
- ✓Everything in Analyst
- ✓API ingestion endpoint
- ✓Spend-threshold webhooks
- ✓Priority support
Start measuring what you actually spend.
Sign up with your Gmail, add your first AI platform, and log your first session in under two minutes.
Create free account