Cheap AI API. Compare the cost of your workload.

AI API prices per million tokens

USD per 1M tokens. This is the Somnus published rate card, not a verified comparison of every provider. Ordered by a 3:1 input/output blend; a different workload can change the order. Effective rates describe credit consumption, not a separate discounted token charge.

What does 1,000 tasks actually cost?

List usage cost = requests × (input tokens × input rate + output tokens × output rate) ÷ 1,000,000. Effective paid cost at standard 2× credit = list usage cost ÷ 2.

These are calculated examples, not measured benchmarks. Chat replies assume 1,500 input and 500 output tokens each; document summaries assume 8,000 input and 400 output; coding steps assume 4,000 input and 3,000 output. An agent task may contain many steps, so 1,000 steps does not mean 1,000 completed projects.

Worked workload estimates

All rows assume uncached text tokens and no retries or extra tool calls. Reasoning tokens, repeated conversation history, and additional requests can increase cost. These figures exclude tax and are usage estimates, not the minimum payment required at checkout.

Why credit and checkout totals are different

Usage is deducted from your prepaid balance at the list rate. With standard 2× credit, a $5 top-up adds $10 of API credit. An estimated $0.33 of effective usage does not mean you can buy a $0.33 top-up: the minimum top-up is $5, and the remaining balance can fund later requests.

The credit multiplier is applied to the pre-tax Stripe subtotal after discounts, not to tax. Standard top-ups receive 2×; $50 receives 2.2×, $100 receives 2.35×, and $200 receives 2.5×. Your first-ever paid top-up receives at least 2.5×. The estimates above deliberately use only the standard 2× rate.

What your top-up adds

No subscription or automatic top-up. Requests stop when available credit is exhausted. Check your checkout total and available balance before running a large job.

Choose by successful task, not just token price

• Run a representative sample on a lower-priced model and a stronger model. Record acceptable answers, input/output tokens, retries, and total spend.
• For document summaries, compare input rates first. For long writing or code output, output rates can matter more.
• Include conversation history and tool results in input estimates. Cap output length and the number of agent steps.
• Do not assume equal capability, speed, context length, cache support, or availability from a price table. Validate the features your application needs.

Check token usage with the OpenAI Python SDK

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["SOMNUS_API_KEY"],
    base_url="https://somnus.world/api/public/gateway/v1",
)

resp = client.chat.completions.create(
    model="claude-sonnet-5",
    messages=[
      {"role": "user", "content": "Summarise this in one sentence: API costs depend on input and output tokens."}
    ],
)
print(resp.choices[0].message.content)
if resp.usage:
    print("Input tokens:", resp.usage.prompt_tokens)
    print("Output tokens:", resp.usage.completion_tokens)

Install with pip install openai and set SOMNUS_API_KEY in your server environment. This example retains claude-sonnet-5. It is not evidence of a completed live request; compare the returned usage with your account’s usage record, and never expose your key in browser code.

Compare direct providers fairly

The sources below include the Somnus rate card and official OpenAI, Anthropic, Google, and DeepSeek pricing references. Check the exact model version, uncached versus cached tokens, batch discounts, fees, tax, and current availability before comparing. Their current prices have not been independently verified for this guide, so no third-party price or market-wide cheapest claim is presented.

Somnus is an independent service, not an official provider product. A model label or response alone does not prove its identity; read the Somnus verification page and test a small workload before committing to a larger top-up.

Questions

Related