Guide
Choosing a model: price, context and capabilities
With several models on one key, how do you pick? A practical way to choose by what the task needs, and to check you chose well.
· 2 min read · Usefully team
One API key gives you models from several providers, and switching between them is a change of model ID. That makes it cheap to pick the right model for each job instead of one model for everything.
Start from what the task needs
- Context window: how much text fits in one request. Long documents, big codebases and long conversations need a large one.
- Capabilities: tool calling for agents, image input for screenshots and photos, JSON mode for structured output. The models page filters by each.
- Price: input and output tokens are priced separately, and output usually costs more. A task with short prompts and long answers is driven by the output price.
- Speed: smaller models generally answer faster, which matters for chat and autocomplete.
Use the smallest model that does the job well
Classification, extraction, routing and short summaries are often handled well by smaller, cheaper models. Save the larger models for hard reasoning, long-form writing and complex code. Many apps use both: a small model for most requests and a larger one when it matters.
Check with your own prompts
Benchmarks are a starting point, not an answer. Take twenty real prompts from your app, run them through two or three candidate models in the playground or a short script, and compare the answers and the cost in Logs. Because the API and the key are the same, the comparison is a loop over model IDs.
for model in ["anthropic/claude-haiku-5-5", "anthropic/claude-sonnet-5-5"]:
reply = client.chat.completions.create(model=model, messages=messages)
print(model, reply.usage.total_tokens, reply.choices[0].message.content[:200])Let routing handle reliability
Once you've picked a model, you don't need to pick a provider. If a model is served by more than one, Usefully routes around failures automatically. See how automatic fallback works.