PromptMax supports Google Gemini, Anthropic Claude, and OpenAI GPT models. That's good news — but it means you have a decision to make every time you start a new job. Not all models are equal, and the wrong choice costs you either money or quality.
This guide cuts through the marketing copy. Here's a practical breakdown of when to use each model family, what the cost difference actually looks like, and a reference table you can use to make the call quickly.
The short version
If you need volume, speed, and predictable structured output — reach for Gemini. If you need nuanced writing, complex reasoning, or tone-sensitive tasks — reach for Claude. If your team already uses ChatGPT and output consistency across tools matters — reach for GPT. If you're not sure, start with Gemini Flash and switch only if the output quality isn't good enough.
What each model family is actually good at
Google Gemini (Gemini 3.1 Pro, 2.5 Pro, Flash, Flash Lite) has been built with throughput and structured output in mind. It handles high-volume, uniform tasks well — tasks where every row looks similar, the instruction is precise, and the output format is clearly defined. SEO meta tags, product descriptions, classification, sentiment scoring, ad copy. Gemini 3.1 Pro is the most capable in the family; Flash remains the workhorse for volume tasks.
Anthropic Claude (Claude Opus 4.7, Sonnet 4.6, Haiku 4.5) tends to produce more natural, varied prose. It follows nuanced instructions reliably and handles tasks where the "right" answer isn't a format — it's a judgement. Personalised emails, job descriptions, editorial rewrites, tasks where tone matters and you'd notice if two adjacent rows read identically.
OpenAI GPT (GPT 5.5, 5.4 mini, 5.4 nano) is a strong all-rounder with a familiar output style. If your team is used to ChatGPT outputs and consistency between your batch pipeline and interactive tooling matters, GPT is the natural choice. GPT 5.5 is the most capable; 5.4 mini and 5.4 nano are optimised for cost at scale.
None is universally better. They make different trade-offs.
How the pricing compares
PromptMax charges based on the tokens your job uses, with a 50% markup over the batch API rates. Here's what each tier costs per million tokens (input / output):
| Model | Provider | Input / 1M tokens | Output / 1M tokens | Best for |
|---|---|---|---|---|
| Gemini 2.5 Flash Lite | £0.06 | £0.18 | Very high volume, simple tasks | |
| GPT 5.4 nano | OpenAI | £0.12 | £0.74 | High volume, GPT-style output |
| Gemini 2.5 Flash | £0.19 | £0.56 | High volume, moderate complexity | |
| GPT 5.4 mini | OpenAI | £0.44 | £2.67 | Mid-range GPT tasks |
| Claude Haiku 4.5 | Anthropic | £0.59 | £2.96 | High volume Claude tasks |
| Gemini 2.5 Pro | £0.94 | £3.56 | Complex structured tasks | |
| Gemini 3.1 Pro | £1.19 | £7.11 | Latest Gemini, complex reasoning | |
| Claude Sonnet 4.6 | Anthropic | £1.78 | £8.88 | Nuanced writing, reasoning |
| Claude Opus 4.7 | Anthropic | £2.96 | £14.80 | Highest-stakes writing tasks |
| GPT 5.5 | OpenAI | £2.96 | £17.78 | Highest-capability GPT tasks |
Flash Lite and GPT 5.4 nano sit at the bottom of the cost curve — fractions of a penny per row for tasks that work on them. Claude Haiku is the entry point for Claude's writing quality. Sonnet is where most people land when they need Claude's prose without paying Opus rates. Gemini 3.1 Pro and GPT 5.5 are the flagship options in their families.
Timing
Gemini jobs run on Vertex AI's batch infrastructure, which typically completes in 20 minutes to a few hours. Anthropic's batch API runs on a similar background model and typically takes one to four hours. OpenAI's batch API has a 24-hour completion window, though most jobs complete within a few hours.
None of these are instant, and all are background processes. For most use cases — starting a job and coming back to the results — the provider timing difference isn't a deciding factor.
Quick reference: which model for which task
| Task type | Recommended | Why |
|---|---|---|
| SEO meta tags | Gemini Flash | Structured output, uniform input, high volume |
| Product descriptions (factual) | Gemini Flash | Reliable format adherence, cost-efficient at scale |
| Product descriptions (brand voice) | Claude Sonnet | Tone consistency, varied prose that doesn't repeat itself |
| Ad copy / headlines | Gemini Flash | Character limit compliance, volume, speed |
| Survey coding / classification | Gemini Pro | Structured multi-field output, handles varied response length |
| Cold email personalisation | Claude Haiku | Natural sentence construction, avoids generic phrasing |
| Job descriptions | Claude Sonnet | Consistent tone, follows nuanced style guidelines |
| Lead qualification notes | Gemini Flash | Structured scoring + short summary, fast and cheap |
| Content localisation | Claude Sonnet | Cultural nuance, idiomatic adaptation, not just translation |
| Review sentiment analysis | Gemini Flash | Classification task — high volume, structured output |
| Data cleaning / standardisation | GPT 5.4 mini | Reliable structured output, strong instruction following |
| Document summarisation | Either Pro | Depends on document complexity and required summary quality |
The tier decision within each family
Once you've picked a provider, you still need to choose the tier. The rule of thumb is simple: start cheap, upgrade only if the quality isn't good enough.
Within Gemini: Flash Lite handles very short, simple inputs well — one-line inputs, single-field outputs. Flash is the workhorse for most tasks. Gemini 3.1 Pro and 2.5 Pro are worth paying for when inputs vary significantly in length and complexity, or when the task requires genuine reasoning across multiple fields. Try 3.1 Pro if you want top Gemini quality — it's slightly cheaper per input token than 2.5 Pro.
Within Claude: Haiku is fast and cost-efficient — it works well where Claude's writing quality matters but the reasoning load per row is low (cold email openers, short rewrites). Sonnet is the right default for anything requiring consistent tone or complex instruction-following. Opus is for highest-stakes work where you can't afford mediocre output — long-form editorial, complex analysis.
Within GPT: 5.4 nano is the entry point — fast and very cheap, good for structured tasks with predictable formats. 5.4 mini is the sweet spot for most GPT jobs. GPT 5.5 is the flagship and worth paying for when output quality on complex, open-ended tasks genuinely matters.
A practical workflow
If you're uncertain, this approach saves both time and credit:
- Run 20–30 rows on Gemini Flash first. It's the cheapest option that produces usable output for most tasks.
- Read the results honestly. Is the output format correct? Does it follow your instructions? Does the writing sound like what you wanted?
- If the format is wrong: improve your prompt, not the model. Model quality rarely fixes a vague instruction.
- If the format is right but the writing quality is off: try Gemini Pro or GPT 5.4 mini. If it's still not there, switch to Claude Haiku or Sonnet.
- Re-run the sample on the upgraded model before committing the full dataset.
The cost of a 30-row sample is negligible on any model. The cost of running 5,000 rows on Claude Opus or GPT 5.5 when Gemini Flash would have done the job is not.