AI model picker
Tell it the job, how often it runs, and how long the prompts are. It returns three API models to test, with each vendor's own reason, the monthly cost at list price, and the test to run before you choose.
The default answers
These are the picker's shortlists for 10,000 tasks a month with a 5,000-token prompt. Prices come from the same file as our AI feature cost calculator, checked 2026-10-13 on the OpenAI, Anthropic, and Google pricing pages. To see any models side by side, with context windows and retirement dates, use the AI model compare tool.
| Job | Test first | Then | Then | How to test |
|---|---|---|---|---|
| Draft text: emails, product copy, summaries | Claude Sonnet 5.5 ($208) | GPT-6.1 Sol ($160) | Gemini 3.8 Flash (2026 price) ($60.00) | Write 20 real requests. Read the drafts side by side with the model names hidden, and pick the one you would send. |
| Extract or classify: invoices, forms, tickets | Claude Haiku 5.5 ($8.45) | GPT-6 Luna ($6.50) | Gemini 3.5 Flash-Lite ($22.50) | Run 50 real documents where you already know the answer. Count the fields that come back wrong. |
| Answer customers in a chat | Claude Sonnet 5.5 ($169) | GPT-6.1 Sol ($130) | Claude Haiku 5.5 ($8.45) | Replay 30 past tickets. Check every fact against your help docs, and check the tone. |
| Agent that uses tools and takes actions | Claude Opus 5.5 ($1,040) | GPT-6.1 Sol ($400) | Claude Sonnet 5.5 ($520) | Run 10 real tasks end to end. Count wrong tool calls and wrong actions, not just wrong final answers. |
| Read long documents: contracts, reports, manuals | Claude Sonnet 5.5 ($234) | Gemini 3.8 Flash (2026 price) ($67.50) | GPT-6.1 Sol ($180) | Ask 10 questions whose answers sit deep in the document, and check each quote against the source. |
| Hard reasoning, research, or long agent runs | Claude Fable 5.1 ($3,250) | GPT-6 Astra ($2,500) | Claude Opus 5.5 ($1,300) | Run the same tests on the mid-tier model too. Move up only if it fails them. Anthropic: "Tuning effort is often a better lever than switching models." |
How the picker decides
- Each job has a default shortlist of three models from at least two vendors, so you can test across vendors and keep a fallback.
- If replies must be fast, Claude Haiku 5.5 and GPT-6 Luna move to the front.
- Very long prompts drop Claude Haiku 5.5, which costs five times more above 100,000 tokens, and warn about OpenAI's 272,000-token step.
- Costs are list price, standard tier, no caching. Claude rows add 30% for its newer tokenizer.
- The shortlists are our judgement. They are not benchmark results. The test step is what decides.
Questions clients ask
Short answers, no sales language. If yours isn't here, ask an engineer directly.
Ask us anythingWhich AI model should my business use?
For most product features, start by testing a mid-tier model such as Claude Sonnet 5.5 or GPT-6.1 Sol, both $2 per million input tokens and $10 per million output tokens. For extraction and classification at volume, start with Claude Haiku 5.5 or GPT-6 Luna at $0.10 and $0.50. Move to a top model only if the cheaper one fails your own tests.
Is the most expensive model always the best choice?
No. The top models cost about 5 times the mid tier per token. Anthropic's own guide says "Tuning effort is often a better lever than switching models." Test the cheaper model on 20 to 50 real inputs first.
Should my staff use ChatGPT or Claude, or should we use the API?
They are different purchases. A staff chat plan such as ChatGPT Business or Claude Team costs $20 per seat per month billed annually, and the vendor picks the model. The API is for putting a model inside your own software, and you pay per token.
Does the tool store my answers?
No. It runs in your browser and nothing you pick is sent anywhere.
Want the test run for you?
Send us 20 real inputs. An engineer runs them through the shortlist and replies with the results and a written scope.