Official OpenAI API prices, verified 18 September 2026. Enter your expected monthly volume to estimate standard, cached, and batch usage. Tool calls and cloud-provider prices are not included.
GPT-6 Astra text pricing
USD / 1M tokens
When it applies
Standard input
$10.00
New input tokens in a standard request.
Cached input
$1.00
Input tokens served from prompt cache.
Cache write
$12.50
New cache-write tokens; 1.25× standard input.
Output
$50.00
Generated tokens, including output associated with reasoning work.
Batch / Flex
50% of standard
Eligible asynchronous or flexible processing.
Fast
2× applicable rate
Higher-speed processing; compatibility and regional restrictions apply.
Long context
2× input/cache · 1.5× output
Whole requests with more than 272K input tokens.
GPT 6 Astra pricing, price, and cost
“GPT Astra cost,” “GPT 6 Astra cost,” “GPT 6 Astra price,” and “GPT 6 Astra pricing” all require the same inputs: the current provider rate card, input/output volume, cache behavior, and any tool or long-context charges. This calculator is a transparent planning estimate, not a quoted invoice or a promise of account eligibility.
How the estimate works
Input cost separates your cached and non-cached shares, then adds output cost. The selected Batch workload percentage receives a 50% discount across this simplified estimate. For an audit-ready forecast, use actual API usage, include cache-write and tool-call charges, and isolate any request whose prompt exceeds 272K tokens because the long-context price applies to the full request.
Planning guidance
Measure a representative week before forecasting a month.
Track reasoning effort separately: higher effort may change output and task completion economics.
Put stable system prompts and long reusable context at the front of the request to make caching practical.
Set per-project spend limits and alerts before enabling autonomous tool use.
Answers
Frequently asked questions
Does this calculator include web search or computer-use charges?
No. It estimates text-token charges only. Tool-specific fees must be taken from the current provider pricing page and added to your model forecast.
Why is my invoice different from this estimate?
Actual usage can include cache writes, tool calls, long-context premiums, taxes, regional pricing, different service tiers, and changes in token volume caused by prompts or reasoning effort.
What is the lowest-cost way to use Astra?
Use the least reasoning effort that passes your evaluation, cache stable prefixes, batch work that does not need immediate delivery, and route routine workloads to a lower-cost model after testing.