LLM APIs
9 articles — every claim carries a date and a source.
Coding Plan vs API: Stop Paying for Tokens Twice
A $10 coding-plan budget bought ~1.9B tokens/month at API prices, and our mid-tier agent plan moved 7–10B tokens in a month. The same volume through raw APIs is a car-payment bill. When the subscription model quietly wins — and the two cases where it doesn't.
4 min read
LLM API Invoice Traps: Thinking Tokens, Cache Charges, Context Multipliers
All three majors officially bill reasoning tokens you never see, OpenAI quietly added a cache write charge, Gemini doubles input price past 200K, and Anthropic adds 10% for US-region inference. Six verified fine-print traps that add 30–200% to a naïve token estimate.
5 min read
Z.ai GLM Coding Plan: Credits → Tokens Math, Plus How to Read Other Plans' Quotas
Z.ai is the only provider that publishes its credit math — here is every plan (Lite/Pro/Max) converted to weekly and monthly token volumes, plus why Claude's '5x' and Google's tiers tell you almost nothing.
6 min read
Coding Plan vs Pay-Per-Token API: Where the Break-Even Sits (November 2026)
Subscriptions only beat API prices above a certain monthly volume. The break-even math on real token mixes — why ~300M tokens/month is roughly where plans start winning, why the OpenRouter anomaly cuts GLM list prices 20×, and when pay-as-you-go is simply the right answer.
4 min read
My Month with the Z.ai GLM Coding Plan (Pro, $64/mo): 9.6 Billion Tokens
Personal experience: 30 days on a real Z.ai GLM Pro subscription under a 10–12h/day agent workload — 9.63B tokens, 92.5% cache hit, 473K credits, nearly 2× the official quota. How much was bonuses, what the same volume costs on API prices, and why the number is a ceiling, not a promise.
4 min read
How to Burn Your AI Coding Tokens 2–5× Slower: Off-Peak, Prompt Caching & Sub-Agents
The same $80 coding plan can deliver 1.2B or 7.5B tokens a month depending on how you work. The four levers: off-peak discipline, context caching (30–180K corridor), orchestrator + cheap sub-agents, and the PLAN.md relay between sessions.
8 min read
The 10 Most-Used Models on OpenRouter, Priced: The $0.006 Input Champion and the $20 Output Flagships
Live pull from OpenRouter's API (Nov 11, 2026): the ten models eating the world's token traffic — DeepSeek V4.1 Flash at 53.8T tokens/week leads — with input/cached/output per-1M prices, what each tier is for, and the Wafer anomaly that hosts GLM-5.3 20× cheaper than its own API.
4 min read
The Output Token Trap: Thinking Loops Burn $20 Before Writing 5 Lines of Code
Output tokens cost 3–5x input tokens. Unbounded agent 'thinking' loops exploit that asymmetry — here's the math and the circuit breakers that stop it.
3 min read
The Prompt Caching Trap: Why Your AI Agent Pays Full Price Every Turn
Prompt caching cuts repeated-context cost by ~90% — but only if the prefix stays byte-identical. How agents silently break their own cache and double your bill.
3 min read