Which LLM
Should I Use?
There are dozens of models — GPT-4o, Claude, Gemini, Llama, Mistral, DeepSeek — and picking the wrong one costs you money, latency, or capability. Answer four quick questions and get a ranked recommendation with reasons.
What are you building?
Pick the primary task your AI feature needs to handle.
What matters most?
This weights the scoring heavily in that direction.
Any data / privacy constraints?
EU or self-host requirements narrow the model list.
Expected request volume?
High volume boosts the weight given to cost and speed.
Recommended
GPT-4o mini
Top pick: solid all-around option for chat / support with balanced priority.
Runners-up
Gemini 1.5 Flash
Good alternative: solid all-around option for chat / support with balanced priority.
Llama 3.x (open)
Good alternative: solid all-around option for chat / support with balanced priority.
Claude Haiku 3.5
Good alternative: solid all-around option for chat / support with balanced priority.
Know your pick? Price it out.
Use the AI Cost Calculator to estimate your monthly API bill before you commit to a model.
Not sure which model is right for your specific use case?
Book a Free Call →Guidance only — model capabilities, pricing, and availability change frequently. Scores are approximate heuristics designed to point you in the right direction; always verify with provider documentation and run your own evals before committing to a model in production. Nothing you enter leaves your browser.
There is no single best model — only the best for your job
Founders often default to “use GPT-4o” because it is the most familiar name. But GPT-4o mini, Claude Haiku, or Gemini Flash can handle the same task at a tenth of the cost — and for high-volume features, that difference compounds to thousands of dollars a month. On the other end, a complex coding agent that needs to reason over a large codebase genuinely benefits from Claude Opus or o3, even at premium pricing.
The right question is not “which model is best?” but “which model is best for this task at this volume with these constraints?” Smart teams route different tasks to different models: a cheap fast model for chat, a smarter model for extraction, and a reasoning model only when strictly necessary.
This picker scores models across intelligence, speed, cost, context length, and capability flags to give you a starting point. Once you have a candidate, use the AI Cost Calculator to see what it will cost at your actual usage volume before you ship.
Frequently asked questions
Which LLM is best for coding?
Claude Sonnet 4 and Claude Opus 4 consistently top coding benchmarks in 2024–2025, followed closely by GPT-4o and o3. For complex algorithmic reasoning and debugging, o3's step-by-step approach excels. For everyday coding at lower cost, GPT-4o mini or Claude Haiku 3.5 can handle straightforward tasks. Run the picker above and choose "Coding" + "Smartest" to get a tailored recommendation.
What's the cheapest good LLM?
For high-volume, cost-sensitive workloads, GPT-4o mini and Gemini 1.5 Flash are among the cheapest closed-source options. DeepSeek offers impressive capability at very low cost. For self-hosted zero-marginal-cost inference, Llama 3.x is a strong open-weights option. The right choice depends on your task — use the picker with "Cheapest" priority to get a ranked list.
GPT-4o vs Claude vs Gemini — which is better?
There is no single winner — it depends entirely on your use case. Claude excels at coding and long-context tasks. GPT-4o has the broadest ecosystem and reliable function calling. Gemini 1.5 Pro/Flash stands out for very long documents (up to 2M tokens). For most "balanced" applications, Claude Sonnet 4 and GPT-4o are neck and neck. The picker weights these trade-offs based on your specific task and priorities.
Which LLMs can I self-host or use in the EU without data leaving my control?
For self-hosting, Llama 3.x (Meta) and Mistral models have open weights you can run on your own infrastructure. Mistral Large is also available via Mistral's EU-based API. For EU data residency without self-hosting, Mistral's API is GDPR-friendly. Select "EU / data control" or "Self-host / open weights" in the picker above to filter to these options.
How much will the model cost to run?
Model pricing varies enormously — from under $0.10/1M tokens for Flash-tier models to $75/1M output tokens for Claude Opus 4. Once you know which model you want, use the AI Cost Calculator to estimate your monthly bill from your expected token usage and request volume.
Need help choosing and implementing the right model?
I help founders pick the right model stack, build routing logic, evals, and cost controls so your AI feature ships fast and stays cheap.