GPT-6 Astra cost model

Updated September 23, 2026 · first published September 23, 2026

GPT-6 Astra is the top model in OpenAI's GPT-6 family. It launched on 3 September 2026, three weeks before GPT-6 Sol and Luna. It costs 5× GPT-6 Sol per token, and on public benchmarks most of its effort settings cost more than a competitor that scores higher.

GPT-6 Astra prices

TierInputCached inputOutput
Standard$10.00$1.00$50.00
Batch or flex$5.00$0.50$25.00
Fast mode$20.00$2.00$100.00
Long context (over 272K input)$20.00$2.00$75.00

Per 1M tokens, from OpenAI's pricing page. A cache write costs $12.50. Astra has a 1,050,000-token context window, 128,000 max output tokens and an April 2026 knowledge cutoff.

Astra always reasons. Unlike Sol and Luna it has no none effort setting; the choices are low, medium, high, xhigh and max.

Against other premium models

ModelInputCachedOutput
GPT-6 Astra$10.00$1.00$50.00
Claude Fable 5.1$10.00$0.25$50.00
Claude Opus 5.5$4.00$0.20$20.00
GPT-6 Sol$2.00$0.20$10.00

Astra and Claude Fable 5.1 have the same list price, but Fable 5.1 charges a quarter as much for cached input. On a cache-heavy agent that difference is most of the input bill.

Cost per task at each effort level

Artificial Analysis Intelligence Index (v4.3.2):

EffortIndex scoreCost per task
low45.8$0.82
medium49.6$1.54
high50.9$1.73
xhigh52.4$2.31
max52.7$3.26

Only Astra at low effort sits on the price-performance frontier. At 45.8 for $0.82, it fills the gap between Sol at xhigh (44.1, $0.53) and Sol at max (47.5, $1.06). It is also frugal: about 4,400 output tokens per task at low, against 27,200 at max.

From medium upward, Claude Opus 5.5 beats it on both axes. Opus 5.5 at medium scores 51.2 for $1.34, higher than Astra at medium or high and cheaper than either. Opus 5.5 at high scores 53.6 for $1.82, higher than Astra at max for 44% less per task. Full cross-model numbers are in real cost per task.

That is a benchmark result, not a verdict on your workload. It does mean Astra above low effort has to earn its place on your own tasks before you pay for it.

What an agent step costs

A step with 200K input tokens, 90% cached, and 20K output tokens costs $1.38 on Astra: 5× GPT-6 Sol ($0.276) and 2.7× Claude Opus 5.5 ($0.516). At the long-context rate the numbers climb fast: the same step at 270K input is $1.51, at 300K it is $2.64.

Fast mode doubles everything

Fast mode (formerly priority processing, renamed 30 July 2026) costs $20 input and $100 output per 1M tokens. Track service_tier on every request. Both "priority" and "fast" select it, so a dashboard filtering on one value will miss spend under the other.

When Astra is worth it

Sources: OpenAI GPT-6 Astra model page, OpenAI pricing, Artificial Analysis. Prices read 23 September 2026.

Related


Want this applied to your own LLM spend? FinOps LLM runs a free audit of your AI costs and shows where the savings are. Book free audit →

Back to research