Friday, so Cartoon Nick is talking money. Specifically, the money most people are lighting on fire by using the most expensive AI model for every single task.
Here is what changed in the last two weeks. Z.ai released GLM-5.3 Flash, an MIT-licensed open model at roughly 7 to 25 cents per million tokens. Alibaba's Qwen 3.8 Flash is in the same neighborhood. Inception's Mercury 2.5 Preview is a diffusion language model that is fast enough to feel instant and priced at under a dollar per million. Meanwhile the frontier models run $10 to $50 per million.
That is a 100x to 500x price gap. And for a lot of tasks, you will not notice the difference.
AI at Work Tip #59: Match the model to the job, not to your ego
The tiering Nick uses in his own shop:
Budget model (GLM, Qwen Flash, Gemini Flash): reformatting, summarizing, tagging, categorizing, first drafts of social captions, pulling data out of PDFs, rewriting for reading level.
Mid model (Sonnet-class, Gemini Pro-class): client emails, blog drafts, code you will actually ship, anything a customer reads.
Frontier model (Fable, GPT flagship): strategy, legal or contract review prep, complex multi-file code, anything where a wrong answer costs real money.
Copy-paste prompt to have the AI sort your own workload:
Here is a list of tasks I do with AI each week: [paste list]. Sort them into three tiers: Budget (a cheap fast model is fine), Mid (needs solid quality), Frontier (mistakes are expensive). For each, give one sentence on why. Then estimate what percent of my total AI usage could move to the Budget tier.Most people find 60 to 70 percent of what they do belongs in the cheap tier.
AI at Work Tip #60: Use "contributor" and cache pricing before you pay list price
Two pricing quirks worth knowing this week. Meta's Muse Spark 1.3 has a contributor tier at 10 to 20 cents per million if you are willing to share usage data. For non-sensitive work like drafting social posts, that is a real discount. And Anthropic dropped cache read pricing by 75 percent, which means if you set up a system prompt once and reuse it (your brand voice, your services, your rules) every call after the first is dramatically cheaper.
The habit: build one reusable "brand brief" and paste it at the top of every session, or set it as a project or custom instruction.
BRAND BRIEF
Business: [name], [what you do], [city]
Voice: [3 adjectives]. Write like a human. No em dashes, no corporate speak.
Never say: [words to ban]
Always: [CTAs, URLs, formats]
Audience: [who]Fifty tips in, that brief is the single highest-leverage thing you can build. It makes every cheap model sound like you.
What's hot right now
Open-weight models are the real story of fall 2026. Mistral, DeepSeek, Qwen, and now GLM give you data sovereignty and no vendor lock-in, and the quality gap to the big names is closing every month. If you run anything customer-facing, this is the year to stop assuming premium means better.
Recurring reminder from your web guy
Every dollar you save on AI is a dollar you can put toward the thing that actually makes you money: a website that ranks, loads fast, and converts. Care plans, hosting, and AI integration all live at webguynick.com.
Frequently asked questions
Are cheap AI models good enough for business use?
For summarizing, formatting, tagging, and first drafts, yes. Reserve premium models for customer-facing or high-stakes work.
What is an open-weight AI model?
A model whose weights are published so you can run it yourself or through any provider, avoiding lock-in to one company.
