How much does a token cost?

Effectively nothing per unit, and everything in aggregate. At $2 per million input tokens, one token costs $0.000002; at $10 per million output, $0.00001. Both round to zero on any receipt. The volume, multiplied across production traffic, is the entire AI budget, which is why token cost is a volume problem wearing a price tag.

What is a token in an LLM?

The chunk of text a model reads or writes: a word, part of a word, punctuation, or whitespace. English averages about four characters per token, and every provider counts them slightly differently. Providers meter and bill by tokens because model compute scales directly with them.

How many tokens is 1,000 words?

Roughly 1,300 tokens in English, though it varies by tokenizer and content: code, numbers, and non-English text typically tokenize less efficiently. For budgeting, 1.3 tokens per word is a safe English-language rule of thumb; verify with a tokenizer tool for anything precision-sensitive.

Why did my token bill jump when usage didn't?

The usual suspects, in order: a reasoning mode switched on (hidden thinking bills as output), cache hit rates degraded after a prompt change, context or history growth inflated input per call, or an agent workflow added steps. Per-feature metering finds the culprit in minutes; invoice totals never will.

Are cached tokens free?

No, discounted. Cache reads bill at a fraction of the input rate, 10% on Anthropic, with newer OpenAI flagships matching that level, and cache writes cost a premium over standard input. The economics only pay off when prompts are structured so caches actually hit, which is the most commonly failed step.

What is token metering in practice?

Counting every billed token category per call (input, output, cached reads and writes, and reasoning tokens) and attributing the counts to the feature, team, or customer that generated them. Providers meter at the API-key level by default; production-grade metering allocates below that, to the units the business recognizes.

Is a cheaper cost per token always cheaper overall?

No. Tokenizers differ, so a cheaper rate on a denser tokenizer can cost more per sentence. Verbose models spend more of the expensive output side. And a cheap model that needs retries or escalation costs more per completed task. Compare models on cost per task, never on rate alone.

You searched for AI Cost Optimization | Page 3 of 75

Blog

LLM token cost: pricing per token explained

LLM token cost is the per-token price a provider charges to read or write text, quoted per million tokens with output priced ~5x input: the unit every AI invoice is denominated in, and a poor unit for deciding anything.

Blog

AI usage tracking: Monitor spend by team, feature & model

AI usage tracking measures who and what consumes AI across your company, by team, feature, and model, then converts usage into spend and cost per unit of work, putting names on the totals provider consoles leave anonymous.

Blog

AI gross margin: how AI spend hits SaaS profitability

AI gross margin is what's left of SaaS profitability once inference, routing, and AI infrastructure land in cost of revenue: a real, measurable compression that's manageable for companies that can see their cost to serve.

Blog

Codex vs. Claude Code in 2026: cost, capability, and which one your team should pick

A finance-first comparison of OpenAI Codex and Claude Code on real cost per developer, capability, and AI ROI, with a clear framework for deciding which coding agent your team should run.

Blog

AI coding tools cost compared: Copilot, Cursor, Claude Code & Codex

GitHub Copilot, Cursor, Claude Code, and Codex compared on real cost per developer, from $0 to $200 a month at list, and the $100 to $250 they actually reach once agents run all day.

Blog

Grok pricing in 2026: SuperGrok plans, API token costs, and what they mean for your AI budget

Grok pricing splits into consumer SuperGrok plans and per-token API rates that have fallen fast in 2026, with add-on meters most guides skip and a churn rate that turns any static AI budget into a guess.

Blog

Google Cloud Functions is now Cloud Run functions: what it is, and what it really costs

Google Cloud Functions became Cloud Run functions in 2024. Here's what the product is, how its per-request pricing actually works, and why AI workloads raise the stakes on every invoice.

Blog

AI cost governance: policies to control AI spend

AI cost governance turns unpredictable AI spend into an answerable question — through enforced budget caps, token quotas, prompt caching, and unified allocation that maps every dollar to a team, feature, or customer.

Blog

Shipped: Cost anomalies and savings recommendations, delivered into ServiceNow

CloudZero continues to ship new features and enhancements at a fast pace. We're publishing these product update posts to keep you up to date on our velocity and how we're adding customer value.

Blog

LLM cost management: a practical guide for teams that own the budget

A practical guide to LLM cost management: the five-level system for allocating, budgeting, and governing model spend so every dollar maps to a feature, team, and outcome.

The AI Cost Optimization Playbook

Finding the ROI of AI: The Finance Perspective