From the blog
Practical LLM cost guides for developers.
Breakdowns for builders shipping AI products.
When Will Your AI Bill Cross Your Budget? The Run-Rate Arithmetic
Two lines of arithmetic tell you mid-month which day your AI spend will pass your budget, and the day a provider spend limit will stop your requests. Here's how to pick the daily rate.
2026-10-31 · 9 min
Read →Model SelectionShould You Switch to a Cheaper LLM? Check Quality First
A cheaper model can be 20 times cheaper per token. Before you switch a feature, check what it's worse at, test it on a slice of traffic, and know when to keep paying.
2026-10-29 · 8 min
Read →Unit EconomicsWhich Customer Is Costing You Money? AI Cost per Customer
Your average AI cost per customer hides the one account that loses you money. Here's how to measure cost per customer, read it, and fix it.
2026-10-28 · 9 min
Read →ArchitectureHow to Choose an AI Gateway for Multiple LLM Providers
Four questions that decide which AI gateway fits: where it runs, how it charges, what happens when a provider fails, and what it tells you about cost. LiteLLM, OpenRouter, Cloudflare AI Gateway and Portkey compared, with each claim linked to the vendor's own page.
2026-10-16 · 9 min
Read →Agentic AIHow to Track AI Coding Agent Costs per Engineer and per Pull Request
Seats are the predictable half of AI coding costs. Agents running on your own API keys are the other half. Worked examples of cost per PR review, per terminal session, per merged PR and per engineer, what OpenAI and Anthropic show you natively, and how to track the rest.
2026-10-16 · 10 min
Read →Pricing MechanicsOpenAI API Cost: What Each Model Costs per Request, per User and per Month
OpenAI API prices per million tokens for GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna, GPT-5.6 and GPT-4o, worked through per request, per user and per month. Plus Batch and Fast pricing, the 272K long-context rate, caching, images and voice.
2026-10-16 · 11 min
Read →Pricing MechanicsHow image and audio models are actually billed
Image and audio models bill in tokens, not per image or per minute. Audio tokens can cost 8x text tokens on the same model. How providers calculate it, where the numbers live in the API response, how most tools get it wrong, and what we can and can't price.
2026-10-06 · 15 min
Read →Cost OptimizationYour OpenAI Bill Doubled. Here's How to Find the Feature That Caused It.
Why an OpenAI bill doubles and how to find the cause today: group cost by day, model and API key in OpenAI's own dashboards, check the seven usual causes, cap spend, then label every request so the next spike names its feature.
2026-09-11 · 10 min
Read →Model SelectionGPT-5.6 Luna Pricing vs GPT-4o Mini: Real Cost per Request
GPT-5.6 Luna pricing is $0.20 per million input tokens and $1.20 per million output. Against GPT-4o Mini it costs 1.5x to 2x more per request. Plus GPT-4o and Gemini 3.6 Flash.
2026-09-11 · 8 min
Read →Cost Optimization15-20% of Your LLM Requests Are Duplicates. You're Paying for All of Them.
Duplicate requests with no caching represent 15-20% of all requests in a typical production app.
2026-09-05 · 7 min
Read →ForecastingStop Being Surprised by Your AI Bill: How to Know Next Month's Cost Today
Spend forecasting with linear regression on 14 days of data lets you act 7 weeks before the problem becomes critical.
2026-08-28 · 7 min
Read →Cost OptimizationReasoning Tokens: The Invisible Line Item in Every New AI Model
GPT-5.6, Claude Opus 5, and Gemini 3 all reason before they answer. That reasoning is billed at the output rate — and most of it you never see.
2026-08-15 · 9 min
Read →Cost OptimizationPrompt Caching: The 90% Discount Most LLM Apps Aren't Using
Every major provider now discounts a repeated prompt prefix by up to 90%. For apps with a large system prompt or fixed context, it's the biggest cost lever available — and the one most teams leave switched off.
2026-08-11 · 8 min
Read →Model SelectionThe Cheapest LLM APIs, Compared (And Why the Ranking Flips)
A current, per-token comparison of the major LLM APIs — and the reason the cheapest option changes depending on whether your workload is input-heavy, output-heavy, or long-context.
2026-08-11 · 9 min
Read →Monitoring & ObservabilityLLM Cost Alerts and Budget Limits: A Setup Guide
A runaway LLM process can spend a month of budget in an hour, and a provider dashboard won't show it until the next day. Here's what to alert on, where to enforce hard limits, and how to make the alerts actionable.
2026-08-11 · 8 min
Read →Cost TrendsLLM Prices Fell 80% Last Year. Your Bill Doubled Anyway. Here's Why.
Inference costs drop ~50% per year. The average production SaaS AI bill doubled in 2025.
2026-08-10 · 7 min
Read →Unit EconomicsWhy Your AI SaaS Has 52% Gross Margins Instead of 80%: The LLM Cost Problem No One Budgets For
Traditional SaaS runs at 70-90% gross margins. AI SaaS runs at 52%.
2026-08-02 · 10 min
Read →ArchitectureThe $40,000 Bill You Don't Know Is Coming: LLM Vendor Lock-In
Three companies needed to migrate away from OpenAI after a pricing change.
2026-07-22 · 9 min
Read →Model SelectionYou're Using GPT-4 for Everything. That's a 30x Mistake for Half Your Features.
The same task costs $0.20 with GPT-3.5 and $6.00 with GPT-4.
2026-07-20 · 8 min
Read →Cost OptimizationYour System Prompt Is Your Biggest LLM Cost Driver. Most Developers Don't Know This.
A 4,000-token system prompt × 100,000 requests = 400 million input tokens.
2026-07-19 · 7 min
Read →Cost OptimizationThe Silent Cost Multiplier in Every Chat Feature: Your Conversation History
A 20-turn conversation can cost 10x what you estimated.
2026-07-10 · 8 min
Read →Agentic AIThe Invisible 40%: How Retry Logic Is Inflating Your LLM Bill Right Now
A real company's retry-on-timeout logic added 40% to their LLM bill without a single person noticing.
2026-07-08 · 8 min
Read →Real Incidents$200 to $6,000 in One Month: The Gemini Bill That Changed How This Founder Thinks About AI Costs
A real founder's Gemini API bill jumped $5,800 in 30 days with no alerts, no feature breakdown, and no warning.
2026-07-06 · 7 min
Read →Setup & IntegrationHow to Track LLM Costs in Production: A Setup Guide for SaaS
A practical guide to instrumenting every LLM call in your SaaS product.
2026-07-04 · 10 min
Read →Model SelectionGemini vs OpenAI API for SaaS Builders: An Honest Cost and Quality Comparison
Google's models are significantly cheaper for many use cases. Here's when to switch.
2026-07-02 · 7 min
Read →Cost OptimizationHow to Calculate Your Real LLM API Cost (The Numbers Most Guides Get Wrong)
Reasoning tokens, retries, context overhead — your real cost is 2–4x the headline rate.
2026-06-30 · 8 min
Read →Agentic AILLM Agent Cost Control: How One Bad Loop Can Cost You $200 in an Hour
Agentic AI workflows can spend 100x your normal hourly rate in a single infinite loop.
2026-06-28 · 7 min
Read →Tool ComparisonsHelicone vs Langfuse in 2026, and When LLMtrack Fits Instead
Helicone went into maintenance mode and Langfuse joined ClickHouse. How the two compare now, and when a cost-only tool is the better pick.
2026-06-26 · 7 min
Read →Unit EconomicsThe Unit Economics of AI Features: Is Your Chatbot Actually Profitable?
Cost per user, cost per conversation, margin per customer — how to calculate whether your AI features are profitable.
2026-06-24 · 8 min
Read →Monitoring & ObservabilityReal-Time vs Delayed LLM Monitoring: Why the 24-Hour Gap Is Costing You
When your LLM monitoring updates once a day, a runaway loop can run for hours before you see it.
2026-06-22 · 6 min
Read →Cost OptimizationHow to Reduce Your OpenAI API Bill by 40–80% Without Touching Your Prompts
Token optimization, caching, and model switching — ranked by impact and effort.
2026-06-20 · 9 min
Read →