2026-08-26
The Best AI Coding Models Right Now, By Job
Coding models run from $0.07 to $5.00 per million input tokens — a 70× spread. A shortlist by task, with verified prices and context lengths.
Notizen von dort, wo die Anbieter zusammentreffen
2026-08-22
The Best AI Video Models Right Now, By Job
A shortlist by task, not a ranking. Prices run from $0.002 to $0.357 per second — a 178× spread. What each tier is actually for.
2026-08-18
The Best AI Image Models Right Now, By Job
Not a ranking — a shortlist by task. The cheapest model that will do each job, from $0.003 bulk generation to precision editing. Verified prices.
2026-08-13
480p, 720p or 1080p? What AI Video Resolution Actually Costs
Resolution is the biggest lever on an AI video bill — and the cheapest provider changes depending on which tier you pick. Verified prices.
2026-08-08
AI Image Generation Pricing Compared
What one generated image actually costs, from $0.003 to $0.045 — across fal.ai, Replicate and Atlas Cloud, with the rounding trap that doubles bills.
2026-08-03
How Much Does AI Actually Cost? Three Worked Examples
What AI really costs for 10,000 product images, a chat feature at 50,000 messages a month, and 200 short videos — with the unit maths shown.
2026-07-28
Why the Same AI Model Costs Different Amounts at Different Providers
The same AI model is often listed at three different prices. Here is what actually varies — billing units, resolution tiers, run-time billing and discounts.
2026-07-21
Was ist multimodale KI? (Und warum sie selten ein Modell meint)
Multimodal heißt zweierlei. Ein Modell, das viele Eingabetypen liest, ist nicht dasselbe wie eines, das sie erzeugt — und fast keines kann beides.
2026-07-13
Was ist Kontextlänge? (Und was sie wirklich füllt)
Kontextlänge ist ein Budget, kein Feature. Was dagegen zählt, wo die realen Grenzen heute liegen, und was passiert, wenn es nicht mehr reicht.
2026-07-06
Was sind KI-Tokens? (Und warum deine Rechnung höher ist als erwartet)
Tokens in klarer Sprache: was als einer zählt, warum ein Chat über 10 Runden das 22-Fache des getippten Textes abrechnet, und warum Ausgabe-Tokens am meisten kosten.