How to choose an AI model

Turn “what is the best model?” into a decision you can actually defend.

Good model selection separates hard constraints from preferences, uses independent quality evidence, checks current operational facts, and makes the tradeoffs explicit.

Install ModelShortlist Browse all guides

Start with the workload and the failure modes that would make a model unusable.

Apply hard constraints such as tools, context, price, creator, or ZDR before ranking softer preferences.

Use independent benchmark evidence to understand quality and performance where model identity is known.

Re-check current catalog, pricing, capabilities, and provider facts because the answer changes over time.

Define hard constraints before looking at rankings

A model that violates a non-negotiable requirement is not a candidate, regardless of benchmark quality.

Minimum context or completion capacity.

Tool/function-calling or structured-output requirements.

Maximum input or output price.

Creator, provider, or privacy requirements such as ZDR.

Use benchmark evidence for quality, not for every operational question

Artificial Analysis can provide independent evidence about intelligence, coding, agentic performance, and related metrics. That evidence is most useful when combined with the workload rather than treated as an automatic final ranking.

Check the current operating facts

Context, supported parameters, pricing, provider availability, and ZDR endpoints can move faster than editorial comparison content. Those facts can change the shortlist even if model quality stays constant.

Ask for an explainable shortlist, not one unexplained winner

A useful result often includes a best-fit option, a best-value option, and a constraint-driven alternative. The host AI should explain why each candidate is in the shortlist and which evidence matters for the recommendation.

Try it in your assistant

Turn the concept into a current decision.

These prompts ask for workload-specific reasoning rather than a permanent ranking. The answer can change as benchmark and operational evidence changes.

“Help me choose a model for this workload. First identify my hard constraints, then shortlist the strongest current options and explain the tradeoffs.”

“Give me a best-fit and best-value recommendation, and separate benchmark evidence from current price, context, and capability facts.”

“Which constraints actually change the answer for this workload, and which are just preferences?”

Keep exploring

Ask the current model market, not yesterday's blog post.

ModelShortlist runs locally, uses your Artificial Analysis and OpenRouter keys, and gives your host AI current evidence for the workload you actually care about.

Open the install configurator