For most small businesses, the right AI model is the one that handles your actual task well at a price you can sustain, not the newest name on the list. Use a larger model for hard reasoning, a smaller one for high-volume sorting and extraction, and test two or three on your own examples.
Model names change quickly, so this article dates every claim. If you want help picking a model for a specific workflow, discuss an automation project. If you have not chosen a task yet, start with what to automate first.
Key facts (all from the vendors’ own pages, accessed 10 November 2026)
- Anthropic says to start with Claude Opus 5.5 for most workloads, and to use Claude Fable 5.2 for demanding reasoning and long-horizon agentic work. It describes Claude Haiku 5.5 as suited to high-volume, latency-sensitive tasks such as classification, extraction and routing. Anthropic models overview.
- OpenAI says to use GPT-6 Astra for complex reasoning and coding, GPT-6.1 Sol to balance intelligence and cost, and GPT-6 Luna for cost-sensitive, high-volume workloads. OpenAI models.
- Google lists Gemini 4 Argon as its flagship reasoning model, and describes Gemini 3.5 Flash-Lite as its fastest, most cost-effective 3.5 model for high-throughput execution. Gemini API models.
- Australian Government guidance advises trialling AI in one or two areas and trying a few products to test accuracy. business.gov.au.
How do the big three describe their models?
Each vendor sorts its models into a top tier, a middle and a small, fast tier. The table groups them that way using each vendor’s own descriptions. The grouping across vendors is our rough reading, not a ranking from the vendors.
| Tier | Anthropic | OpenAI | |
|---|---|---|---|
| Top: hardest problems | Claude Fable 5.2: demanding reasoning and long-horizon agentic work | GPT-6 Astra: complex reasoning and coding | Gemini 4 Argon: flagship reasoning model |
| Middle: everyday work | Claude Opus 5.5 (the vendor’s starting point) and Claude Sonnet 5.5 (speed and intelligence) | GPT-6.1 Sol: balances intelligence and cost | Gemini 3.8 Flash: its most intelligent Flash model |
| Small: volume and speed | Claude Haiku 5.5: classification, extraction and routing | GPT-6 Luna: focused, high-volume tasks | Gemini 3.5 Flash-Lite: high-throughput execution |
Sources: Anthropic, OpenAI and Google, all accessed 10 November 2026.
These pages are written for developers. If you use a chat app or a business plan instead, the models on offer depend on the app and your plan, so check what yours includes.
What do the differences cost?
The price gap between tiers is large, which is why matching the model to the job matters for automations that run many times a day.
As of 10 November 2026, Anthropic lists Claude Fable 5.2 at $8 per million input tokens and $40 per million output tokens, and Claude Haiku 5.5 from $0.10 and $0.50. OpenAI lists GPT-6 Astra at $10 and $50 and GPT-6 Luna at $0.10 and $0.50. These are developer API prices, quoted in the vendors’ pages. They are not the price of a chat subscription, and they can change.
For a handful of emails a day the difference is small. For a workflow that sorts thousands of messages, it can decide whether the automation is worth running. Get a full estimate for your volume before you build; our AI automation cost guide explains what else goes into the total.
Which model suits which job?
This table is our suggestion for where to start, not a vendor claim. Test before you rely on it.
| Job | Try first | Why |
|---|---|---|
| Draft a reply, summary or checklist you will edit | Middle tier | A person checks the output, so you do not need the most expensive option |
| Sort, label or pull details from many messages | Small tier | The vendors position their small models for classification and extraction |
| Analyse a complicated problem or plan a multi-step task | Top tier | The vendors describe their largest models for demanding reasoning |
| Anything involving customer personal information | Check the account first | Privacy comes before model quality |
Anthropic, OpenAI and Google each list a context window of about one million tokens for their flagship models (Anthropic’s models overview, OpenAI’s models page and Google’s Gemini 4 Argon card, all accessed 10 November 2026), so context size alone is unlikely to separate them.
How do you test a model on your own work?
A short, honest test beats a leaderboard.
- Pick one task. Use one from your own business, such as replying to a routine enquiry.
- Make a small set of examples. Use fictional or public material. Include easy ones, awkward ones and one that should be refused or passed to a person.
- Run two or three models. Try a middle-tier and a small-tier model, and a top-tier one if the task is hard.
- Score the results. Check accuracy against your source, tone, and whether it invented anything. Note how long each took and what it cost.
- Choose the cheapest model that passes. Move up a tier only if the results fall short.
Australian Government guidance says AI tools can make mistakes and present false information as fact, so check output before you use or share it. Keep a person responsible for each result. business.gov.au, accessed 10 November 2026.
What happens when a model is retired?
Models are replaced and retired, so write down which model each workflow uses.
As of 10 November 2026, Google’s model page says gemini-2.5-flash is deprecated, and that from
1 December 2026 requests to it will automatically route to gemini-3.5-flash-lite. Anthropic’s
page lists retirement as “not sooner than” a date for each current model; for Claude Haiku 5.5
it says not sooner than 7 October 2027. Partner platforms can set their own dates.
For each workflow, record the model name, the date you tested it and who will retest it. When a vendor announces a change, rerun your small test set before you switch.
What about privacy?
Choose the account and settings before you choose the model.
The OAIC says privacy obligations apply to personal information entered into an AI system, and recommends not entering personal information, especially sensitive information, into publicly available generative AI tools. It also says to do due diligence on a commercial product before adopting it. OAIC guidance (published 21 October 2024, updated 17 January 2025). Our guide to Claude for Australian small businesses covers account types and a first trial in more detail.
Do you need AI at all for the job?
Not always. A fixed rule or a saved template does not need a model. Use AI when the input varies and each case needs reading. For a worked example, see how to get AI to draft replies to business emails, read how to write an AI policy, or explore AI automation for small businesses.
Frequently asked questions
Which AI model is best for a small business?
There is no single best model. Match the model to the job: a larger model for hard reasoning, a smaller, cheaper one for high-volume sorting and extraction. Test two or three on your own tasks before you commit.
Do I need the newest or most expensive model?
Not for every task. Each vendor describes its largest model as the one for demanding reasoning and agentic work, and its smallest as the one for high-volume tasks. Start where the vendor’s description fits your job, then move up only if the results fall short.
How often do AI models change?
Often. As of 10 November 2026, Google’s model page says requests to gemini-2.5-flash will be routed to gemini-3.5-flash-lite from 1 December 2026. Keep a note of which model each workflow uses and recheck it on a schedule.
Are the prices on the vendors’ model pages what I will pay?
Not necessarily. The pages list prices for developer API use, per million tokens. Chat apps and business plans are priced separately, so check the plan you would actually use.
Can I put customer information into any of these models?
Check first. The OAIC recommends not entering personal information into publicly available generative AI tools, and says privacy obligations apply to personal information you enter. Review the account type, settings and terms before using real customer data.
Want help matching a model to a real workflow? Discuss an automation project.
