Guides
Diagnose a bill, or plan the next one away
Plain-language breakdowns of the most common reasons an AI API bill jumps — reasoning tokens, cache misses, usage tiers, retried requests, and undisclosed vendor price changes — plus practical strategy guides on cost optimization, vendor lock-in, and running more than one AI vendor at once.
- Diagnosis
Why did my OpenAI bill increase?
The most common, explainable reasons an OpenAI API invoice jumps — and what to check first.
- Diagnosis
Why did my Anthropic bill increase?
The most common, explainable reasons a Claude API invoice jumps — and what to check first.
- Diagnosis
OpenAI 429 error: what it means and how to fix it
A 429 from the OpenAI API can mean three different things — here is how to tell them apart and fix each one.
- Diagnosis
Claude API 429 error: what it means and how to fix it
A 429 from Anthropic can be a plain rate limit or a monthly spend cap — the fix is different for each.
- Diagnosis
Azure OpenAI 429 error: what it means and how to fix it
Azure OpenAI's quota works differently from direct OpenAI or Anthropic — here's what actually causes a 429 and how to clear it.
- Diagnosis
Why did my Gemini API bill increase?
The most common, explainable reasons a Gemini API invoice jumps — and what to check first.
- Diagnosis
Gemini API 429 RESOURCE_EXHAUSTED: what it means and how to fix it
A Gemini 429 always comes back as the same error code — unlike other vendors, working out the fix means figuring out which limit you actually hit.
- Diagnosis
Groq API 429 error: what it means and how to fix it
Groq tells you exactly how close you are to a rate limit before you hit one — most recurring 429s are avoidable once you're reading it.
- Strategy
How GPU cloud billing actually works (Runpod, Modal, and similar)
Per-second, per-GPU-hour billing looks simple until idle time, cold starts, and rounding start showing up on the invoice. Here's what's actually billable.
- Diagnosis
AWS Bedrock ThrottlingException: what it means and how to fix it
Bedrock quota is shared across your whole account and region, and an unset max_tokens can reserve far more of it than you think — here's how to actually fix a ThrottlingException.
- Diagnosis
DeepSeek API 429 error: what it means and how to fix it
DeepSeek doesn't rate-limit by requests-per-minute or tokens-per-minute at all — it caps how many requests can be in flight at once. That changes the fix.
- Strategy
LLM cost optimization: the techniques that actually move the needle
Five concrete levers for cutting AI API spend, and which workloads each one actually fits.
- Strategy
How to reduce OpenAI API costs
The concrete levers for cutting an OpenAI bill — prompt caching, the Batch API, output caps, and model choice.
- Strategy
How to reduce Claude API costs
Cache-read discounts, the Message Batches API, capping extended thinking, and matching workloads to the right Claude tier.
- Strategy
How to avoid AI vendor lock-in
Practical patterns for keeping a real second option, without over-engineering an abstraction layer you don't need yet.
- Strategy
Multi-vendor LLM strategy: a switching-providers checklist
What actually changes when you add, or switch, an LLM vendor — beyond swapping an API key.