Home → Guides → GPT-6 Astra API pricing: per-million rates, the 272K tier, and what real calls cost

GPT-6 Astra API pricing: per-million rates, the 272K tier, and what real calls cost

The list price is easy to find. What it means for a bill is not: on real traffic most of the input arrives from cache, and a small share of oversized requests costs far more than its share of calls.

Official price per million tokens

From OpenAI's published rates for gpt-6-astra:

	Input	Cached input	Output
Up to 272,000 input tokens	$10.00	$1.00	$50.00
Above 272,000 input tokens	$20.00	$2.00	$75.00

Above the threshold the whole request is repriced, not just the tokens past the line. On this platform the standard rate is:

Model	Input / 1M	Output / 1M	Share of list price
gpt-6-astra	$2.00	$10.00	20%

What 422 real calls cost

Every billed gpt-6-astra call that has gone through this gateway, priced at list:

Line	Tokens	Cost at list	Share
Fresh input	11,954,396	$135.04	68%
Cached input	42,697,881	$49.55	25%
Output	255,741	$13.11	7%
Total		$197.70

• 78% of all input tokens were cache hits. They cost $1 per million instead of $10, which is why cached input is a quarter of the bill rather than most of it.
• Output is only 7%. The median call returned 261 tokens against roughly 107,000 tokens of context. This is coding-agent traffic: read a lot, write a little.
• The median call cost $0.23 at list; the average is $0.47, pulled up by a minority of very large requests.
• Customers paid 79.13 credits for these calls — $39.57 at 1 USDT = 2 credits — against $197.70 at OpenAI's list price.

The 272K threshold, and the detail most people miss

Only 25 of the 422 calls (5.9%) went over 272,000 input tokens, yet they account for $45.67 — 23% of the whole bill. Crossing the line doubles the input rate and raises output by half for the entire request:

• 270,000 fresh input + 2,000 output, under the line: $2.70 + $0.10 = $2.80
• 300,000 fresh input + 2,000 output, over the line: $6.00 + $0.15 = $6.15

Eleven percent more input, more than twice the cost. And cached input counts toward the threshold: 22 of those 25 calls had fewer than 272,000 fresh tokens, but crossed the line once cache hits were added — and every token, cached ones included, was billed at the higher tier ($2.00 instead of $1.00 for cached input). If your agent keeps appending history, trimming context to stay under 272K saves more than any model switch.

Why caching decides the bill

A turn that re-reads a 200,000-token conversation, adds 20,000 new tokens and writes 1,000 tokens of output:

• With the history cached: $0.20 + $0.20 + $0.05 = $0.45
• With nothing cached: $2.20 + $0.05 = $2.25

Five times apart for the same work. Caching is automatic, but anything that changes at the start of the prompt every turn — a timestamp, a reshuffled tool list — breaks it. Keep the volatile part at the end.

Using gpt-6-astra here

For Codex, add a provider to ~/.codex/config.toml and export your key as APICLAN_API_KEY:

model = "gpt-6-astra"
model_provider = "apiclan"

[model_providers.apiclan]
name = "APICLAN"
base_url = "https://apiclan.us/v1"
env_key = "APICLAN_API_KEY"
wire_api = "responses"

For the OpenAI SDK or a raw request, the base URL is https://apiclan.us/v1:

curl https://apiclan.us/v1/chat/completions \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-astra","messages":[{"role":"user","content":"hi"}]}'

Top-ups are in USDT, at 1 USDT for 2 credits of balance, and there is no monthly fee. The median call above took 19 seconds end to end, with the first token after about 8 seconds; long generations take minutes, so stream the response.

Is Astra worth it?

Astra lists at five times gpt-6-sol. For routine steps — extraction, classification, short edits — Sol does the same job for much less; the split is covered in GPT-6 Sol vs Astra. Keep Astra for the hard reasoning steps, keep requests under 272K, and let the cache do its work: on the traffic above, that is what kept the real cost at a fraction of the list price.

Start using it

No subscription, no monthly minimum, no sales call. Top up with USDT and spend what you use — 1 USDT gives you 2 credits of API balance.

Read the 30-second quickstart

Language: English · 简体中文 · 繁體中文 · 日本語 · 한국어 · Deutsch · Français · Español · Türkçe · Português · Italiano · Bahasa Indonesia · Tiếng Việt · Bahasa Melayu · Polski · Čeština · فارسی · Русский

Prices on this page are filled in automatically from live billing data. Third-party terms are quoted from that party's own documentation.