Which AI model should a marketing team use?

Default to the mid tier for drafting, summarising and analysis. Use the cheap tier for bulk classification and short rewrites, and reserve the flagship tier for hard reasoning where being wrong is expensive.

How much do AI models cost?

They are priced per million tokens, split between input and output. As of July 2026 in the Claude range, Haiku 4.5 is $1/$5, Sonnet 5 is $2/$10 on introductory pricing through 31 August 2026 then $3/$15, and Opus is $5/$25.

How can I reduce AI costs?

Prompt caching can cut input cost up to 90% when reusing the same context, batch processing typically halves cost for non-urgent work, and choosing the right tier matters more than either.

Do newer AI models cost more per request?

Sometimes, even at the same rate. Newer tokenisers can produce roughly 30% more tokens for identical text, so compare models on a real task rather than on the published per-token rate.

Which AI Model for Your Marketing Team

Most teams pick one AI model and use it for everything, which is how you end up paying flagship rates to rewrite a subject line.

Models are priced per million tokens, split between input (what you send) and output (what comes back). Matching the model to the job is usually a bigger saving than any prompt trick.

The tiers, and what each is for

Every major provider now ships roughly three tiers, and the pattern holds across them.

A cheap fast tier for bulk, simple work: classification, tagging, short rewrites, extracting fields from messy text.
A mid tier that handles most real work: drafting, summarising long documents, analysis.
A flagship tier for genuinely hard reasoning, long multi-step tasks and code.

As a concrete example, in the Claude range Haiku 4.5 runs $1 per million input tokens and $5 output, Sonnet 5 sits at $2 and $10 on introductory pricing through 31 August 2026 before moving to $3 and $15, and the Opus tier is $5 and $25. Verified July 2026, and worth re-checking before you build a budget on it.

The rule that saves the most money

Default to the mid tier. Move down for bulk, move up only when the mid tier visibly fails.

For marketing work specifically, the mid tier handles ad copy variations, email sequences, landing page drafts and campaign analysis without difficulty. Reserve the flagship for things where being wrong is expensive.

Three cost facts that change the maths

Prompt caching can cut input cost by up to 90 percent when you send the same context repeatedly, which is exactly what a brand-guidelines prompt does.

Batch processing typically halves the cost for work that is not time-sensitive. Generating a hundred product descriptions overnight does not need to be interactive.

And a subtle one: newer models often use a different tokeniser that produces more tokens for the same text, in the Claude 5 generation roughly 30 percent more. Per-token price parity is not per-request cost parity, so compare on a real task rather than on the rate card.

What this costs a small Indian agency in practice

Rough order of magnitude: drafting copy for a handful of clients, with caching on and sensible model choice, sits in the low thousands of rupees a month rather than the tens of thousands. The cost that actually hurts is a badly built automation looping a flagship model over a large dataset.

Put a spend limit on the account before you build anything. Every provider offers one, and almost nobody sets it until after the first surprising invoice.

What not to automate

Anything a client will read as your professional opinion, unedited. AI drafts fine and commits confidently to things that are not true, and in this market your credibility is the asset you are actually selling.

Want this built for your business?

We build the funnel, the follow-up automation and the ad campaigns as one system. Fixed published pricing, 14-day build.

Book a free strategy call