Which LLM should I use?
Pick your task, what matters most, and how much text goes into each request. Get a shortlist of Claude, GPT, and Gemini models with estimated costs.
Prices checked against official provider pages on October 20, 2026, context windows on October 21, 2026. Runs in your browser.
Include the document, instructions, and any history you send.
Start with mid-tier models. Tone and accuracy both show to customers, so a mid-tier model is a safer default. Route simple, repetitive questions to a small model to save money.
- Gemini 3.8 FlashGoogle · Mid-tier$3.00per 1,000 requests
Strong quality for the price. The usual default for production work. Uses about under 1% of its context window. Rises to $1.50 / $7.50 on January 1, 2027.
- Claude Sonnet 5.5Anthropic · Mid-tier$8.00per 1,000 requests
Strong quality for the price. The usual default for production work. Uses about under 1% of its context window.
- GPT-6.1 SolOpenAI · Mid-tier$8.00per 1,000 requests
Strong quality for the price. The usual default for production work. Uses about under 1% of its context window.
Tiers reflect each provider’s own labels and price, not benchmark results. Costs assume about 400 output tokens per request at standard prices. Test two or three of these on 20 to 50 of your own examples before you commit.
How to choose a model
- Describe the job. Choose the task that’s closest to your use case, then say whether cost or quality matters more.
- Say how much goes in. Longer inputs rule out models with smaller context windows and can raise the price per request.
- Test the shortlist. Run two or three of the suggestions on real examples. Keep the cheapest one that clears your quality bar.
Choosing a model, explained
- Which LLM should I use?
- It depends on the task, your budget, and how much text goes into each request. Simple, high-volume work such as classifying or extracting data usually runs well on a small, fast model. Everyday drafting, support, and coding suit a mid-tier model. Reserve flagship models for hard reasoning and multi-step agent work.
- How were the tiers decided?
- Each model is grouped by its provider’s own labels and its price: small and fast, mid-tier, or flagship. This is my own grouping, not a benchmark result, so use it to build a shortlist rather than to settle the question.
- Is the most expensive model always the best choice?
- No. A model that is more capable than the task needs costs more without improving the result. A cheaper model that gets the task wrong costs more in retries and review. The only way to know is to test on your own examples.
- How should I test the shortlist?
- Collect 20 to 50 real examples with the answer you would accept. Run each shortlisted model on all of them, score the results the same way, and compare accuracy against cost per request. Retest when a provider releases a new model.
- Why does input size matter?
- Every model has a context window, and longer prompts cost more. Some models also charge higher rates above a certain prompt length. The token counter can tell you how large your documents really are.
- Why are some models left out?
- The shortlist shows current-generation models only and one per provider, so you compare real alternatives instead of three versions of the same one. Older models are still in the cost calculator.
I’m Courtney Stephenson, an AI strategy and implementation leader, and I built this tool. If you’re deciding which models belong in your stack, get in touch.
Want a second opinion on your results?
Book a free 15-minute call and we’ll talk through what this means for your team.