What is the cheapest LLM API for my workload?

Estimated monthly cost — Short assistant requests

Calculated live from current TokenRex prices.

Price the whole request

The cheapest input price is not automatically the cheapest model. Add input, output, cached input where supported and the number of extra requests needed to finish a task.

For short classification, input usually dominates. For writing, output can matter more. For agents, repeated context and tool rounds often matter most.

A practical selection process

• Write 20 to 50 real prompts with expected results.
• Run the cheapest likely model first and record failures.
• Escalate only failed or high-value requests to a stronger model.
• Compare cost per accepted result, not cost per successful HTTP response.

Use live comparisons

The table above uses current enabled TokenRex prices for one example workload. Change the token counts in the cost calculator for your own traffic.

Compare specific models on the Models & Pricing page and head-to-head comparison pages before switching production traffic.

Sources

• Google Search Central: helpful, reliable content

Example

from openai import OpenAI

client = OpenAI(api_key="YOUR_TOKENREX_KEY", base_url="https://tokenrex.ai/api/public")

resp = client.chat.completions.create(
    model="gpt-6-luna",
    messages=[{"role": "user", "content": "Classify this ticket as billing, technical or account"}],
    max_tokens=20,
)
print(resp.choices[0].message.content)

Questions

Is there one permanently cheapest LLM?

No. Prices and suitable models change, and your token mix decides the result.

Does TokenRex automatically switch to the cheapest model?

No. Your application chooses the exact model ID.