GPT-6 Astra
45% less- Input
- $5.5OpenAI standard price
$10 - Cached input
- $0.55OpenAI standard price
$1 - Output
- $27.5OpenAI standard price
$50
Complex reasoning · Temporary hold: $7.48
PAY AS YOU GO · NO SUBSCRIPTION
Cheapest LLM API for text generation.
45% lower token rates.
Compare model rates, estimate a real text request, then switch an OpenAI-compatible client when the math works.
No plans to buy. Top up only what you need.Paid balance stays available for future requests.
Pick your model
USD / 1M tokensCompare all 5 model prices
| Model | Input | Cached | Output |
|---|---|---|---|
| 6 Astra | $5.5 | $0.55 | $27.5 |
| 5.6 Sol | $2.2 | $0.22 | $11 |
| 5.6 Terra | $1.1 | $0.11 | $6.6 |
| 5.6 Luna | $0.11 | $0.011 | $0.66 |
| 5.5 | $2.75 | $0.275 | $16.5 |
Compared with OpenAI’s standard published rates · Sep 9, 2026.
Token rates only. Billed input can include service instructions; supported features differ.
Configure your API key
01 → 0301
Get your free API key Get a prixis key
$3 free balance. No card required.
02
Set your base URL & key
For OpenAI-compatible text clients.
.env
OPENAI_BASE_URL=https://prixis.dev/v1
OPENAI_API_KEY=your_prixis_api_keyUse your own key and select GPT-6 Astra. Full configuration guide →
Works with your existing text workflowPythonTypeScriptcURLOpenAI-compatible
CHEAPEST LLM API COST ESTIMATOR
Estimate your cheapest LLM API request before you sign in.
Enter the token shape of a real prompt. prixis calculates the request cost from uncached input, cached input and output tokens, so you can compare the model before sending production traffic.
REQUEST FLOW
Cheapest LLM API requests still need an exact ledger.
A request reserves the model’s temporary hold first, runs once, then settles from the token usage reported by the completed response.
CHEAPEST LLM API FOR THE JOB
The cheapest LLM API request is
a request you only send once.
Match the task, then the model.
A short extraction does not need the same model as a complex code review. Run a small set of your real prompts before committing.
Watch what comes back.
Output costs more than input. Ask for the format and length you actually need; count reasoning usage in your budget.
Recharge what you need.
Choose your own recharge amount. Your paid balance does not expire, and there is no monthly plan to keep filling.
02 / FROM KEY TO FIRST REQUEST
Change the endpoint for your cheapest LLM API.
Keep your text workflow.
Create a key in your console, set the base URL, and send your first text request.
- 1Create your prixis account.
- 2Copy a new API key once.
- 3Send a request and inspect its usage.
from openai import OpenAI
import os
import uuid
client = OpenAI(
api_key=os.environ["OPENAI_API_KEY"],
base_url=os.environ["OPENAI_BASE_URL"],
max_retries=0,
)
request_id = str(uuid.uuid4())
reply = client.chat.completions.create(
model="gpt-6-astra",
messages=[{"role": "user", "content": "Explain this in one line."}],
extra_headers={"Idempotency-Key": request_id},
)
print(reply.choices[0].message.content)Endpoint support and errors
Text chat
GET /v1/modelsPOST /v1/chat/completions
Use text messages with system, developer, user or assistant roles. Select low, medium or high reasoning effort.
Keep retries safe
Every request requires a unique Idempotency-Key for each logical request. A duplicate key with a different request is rejected. A pending request is never silently sent again.
Know the limits
64 KB maximum message text. No image, audio, tool or structured-output support. No sampling or output-token controls. SSE delivers the completed answer in one message.
Read the status
401 Invalid key. 402 Insufficient balance. 409 Request pending or key conflict. 429 Rate or capacity limit. 503 Service unavailable.
03 / CHEAPEST LLM API MODEL PRICING
Every token. Every rate for your cheapest LLM API.
Pay for actual usage. Compare all five models, with every cheapest LLM API rate in one place.
| Model | InputUncached · / 1M tokens | Cached input/ 1M tokens | Output/ 1M tokens | Temporary holdAvailable balance per request |
|---|---|---|---|---|
GPT-6 AstraDefault Complex reasoning | $5.5 | $0.55 | $27.5 | $7.48Released at settlement |
GPT-5.6 Sol Everyday code & chat | $2.2 | $0.22 | $11 | $2.992Released at settlement |
GPT-5.6 Terra Balanced everyday work | $1.1 | $0.11 | $6.6 | $1.7952Released at settlement |
GPT-5.6 Luna Fast, cost-sensitive tasks | $0.11 | $0.011 | $0.66 | $0.17952Released at settlement |
GPT-5.5 Previous-generation flagship | $2.75 | $0.275 | $16.5 | $4.488Released at settlement |
HOW YOUR CHARGE IS CALCULATED
Three token counts. One exact charge.
(Uncached input × input rate + cached input × cached rate + output × output rate) ÷ 1,000,000
Cached tokens are excluded from uncached input. Billable reasoning tokens count as output. Input can include service instructions.
GPT-6 Astra example
1,000 input tokens, including 400 cached, plus 200 output tokens.
600 × $5.5 + 400 × $0.55 + 200 × $27.5÷ 1,000,000 = $0.00902YOUR BALANCE, BEFORE & AFTER
Start free. Top up on your terms.
New accounts get $3 free. Requests require the temporary hold shown above; GPT-6 Astra needs $7.48 available. GPT-5.6 Luna needs $0.17952.
The final charge uses actual token usage. Unused held funds return after settlement or a confirmed failure. While execution is uncertain, funds stay held. Retry with the same Idempotency-Key to avoid a second charge.
Add $0.01–$9,999.99 when you need it. No subscription. Paid funds do not expire and stay available for future API requests.
Open your free accountWhat is prixis?
prixis is a prepaid LLM API service for text generation, coding and reasoning. You can compare model rates, create an API key, send requests through an OpenAI-compatible chat endpoint, and review each request’s token usage.
Which is the cheapest LLM API for my app?
It depends on the model your task needs, your input-to-output ratio, cache usage, and your model’s token rates. GPT-5.6 Luna has the lowest token rates on prixis. Try your own prompts before choosing: a cheaper request is not a saving if its answer needs to be generated again.
How is each request charged?
Uncached input tokens × input rate, plus cached input tokens × cached rate, plus output tokens × output rate. Divide each listed per-million rate by one million. For GPT-5.6 Sol, 1,000 uncached input tokens and 200 output tokens cost $0.0044. Billing retains nine decimal places in USD; charges are not rounded up to a cent.
Are cached input and reasoning tokens billed?
Cached input is charged at the listed cached-input rate when the response includes that usage. Output usage includes any billable reasoning tokens reported by the model. Input usage can include service instructions in addition to your messages. While execution is uncertain, the balance hold stays pending. Once execution ends, a request without a complete result and valid usage fails with no charge. Estimates never become final bills.
Can I use my existing OpenAI SDK?
Yes, for the documented text chat endpoint. Set your base URL to https://prixis.dev/v1 and use a prixis API key. Images, audio, tools, structured-output schemas, sampling controls and output-token limits are not supported. Streaming returns the completed answer as an SSE message; it is not incremental token streaming.
Is this a good fit for production workloads?
Start with development, coding helpers and workloads that can tolerate a wait. Capacity is shared and rate limits may interrupt requests. prixis does not promise dedicated capacity or an uptime SLA. Keep your own retry policy and use the documented Idempotency-Key header to avoid duplicate charges.
Do paid funds expire?
No. You do not need to buy a plan: top up only what you need, and paid funds stay available for future API requests. Payment reversals from the checkout provider can reduce balance and pause new requests until the account is current.
What happens when my balance runs low?
New requests need enough available balance for the model’s temporary hold, shown with its rates and in the console. The final charge is usually much smaller and unused held funds are released. If the balance is too low, the request stops before inference. Add any amount from $0.01 when you choose; there is no automatic recharge.
MAKE YOUR NEXT REQUEST COUNT.