共有:
Pricing & Rollout

GPT-6 Astra
expands upmarket.
What "half price" really means.

GPT-6 Astra, OpenAI's flagship model launched September 3, has widened its reach to top-tier ChatGPT plans within days. Cross-checking the official announcement against the live pricing page reveals a more layered story than the "roughly half of Sol" headline suggests.

AI Navigate Editorial·2026.09.06·6 min read
Sept 3 Admins enable per org Sept 6 · now Top-tier ChatGPT plans expanding Coming days Business Enterprise full rollout Parallel API AWS Just 3 days after launch, who can access it has already widened
01
The Gap

The gap between "launched"
and "usable" just got short

Frontier model launches usually leave weeks or months between the announcement and general availability. GPT-6 Astra's September 3 launch was itself limited: only organization admins could turn it on. Just three days later, on September 6, it became clear that Astra is expanding to top-tier ChatGPT plans — Plus, Pro, Business, and Enterprise — in sequence.

OpenAI has also published Astra's access terms on its API and pricing documentation, signaling that both the ChatGPT and API tracks are being pushed forward at once. For a frontier model, avoiding the usual "announced, then months of silence" pattern is a meaningful shift in how fast the gap between hype and hands-on access is closing.


02
The Price Sheet

Standard, cached, batch, fast
— four ways to pay

Astra's usage-based pricing isn't a single number. Four distinct rate tiers exist depending on how you call it.

$10 / $50
Standard rate, input/output (per 1M tokens)
$1
Cached input (per 1M tokens)
×0.5
Batch runs at half the standard rate
×2
Fast mode: 2x price for up to 2x speed
$25 Batch $50 Standard $100 Fast Input drops to $1 when cached
FIG. Output price per 1M tokens (USD) across modes. Input pricing follows a separate cache-rate schedule.

Writing to cache costs 1.25x the uncached input rate on top of that. The four-tier structure — standard, cached, batch, and fast — lets a team mix and match, but it also pushes the cost-optimization work onto whoever is implementing it.


03
Read The Fine Print

"Half of Sol" depends entirely
on which price you're reading

The headline framing for this update is "priced at roughly half of Sol." But lining up Astra's standard API rate ($10 input / $50 output per 1M tokens) against GPT-5.6 Sol's current API rate puts Astra at roughly 2.5 times Sol's price on both input and output. Not half — the raw numbers say the opposite.

The mismatch comes from comparing two different things. "Half price" appears to describe Astra's position as a model choice inside top-tier ChatGPT plans — a subscription-value framing. A developer pulling up the raw per-token API rate sees a different number entirely: a price higher than Sol's, not lower. Same model, same announcement, opposite conclusions depending on which figure you read. Conflating the two and assuming "Astra is cheaper" outright can throw off a real API migration cost estimate.

Subscription framingAPI metered framing
Included in a top-tier plan at no extra chargeStandard rate: $10 input / $50 output per 1M tokens
Marketed as better value than SolRoughly 2.5x Sol's price on input and output
Subscribers rarely notice the price deltaProduction traffic feels the full cost directly

04
Who It's For

Who should act now, by role

Rollout is proceeding top-tier-plan-first, so what you can actually do today depends on your role.

Engineers

Benchmark both standard ($10/$50) and Fast mode (2x speed, 2x price) against your own workload before committing production traffic. Routine, non-urgent jobs can move to batch for half the standard rate.

Business leads

If your org is on a top-tier ChatGPT plan, you can try Astra hands-on today without waiting for the API rollout — a useful window to evaluate quality before committing to a bigger technical migration.

Product managers

Lower-tier users don't have access yet. Treat this as a small pilot limited to top-tier subscribers, and use it to gather quality evidence before making a company-wide rollout call.

The headline says half price.
The line item says $10 and $50.
They're answering different questions.


05
What To Do Next

Recommended next steps

01

If Sol is stable in production, don't rush

Rollout is still spreading "over the coming days" and Astra isn't evenly available to everyone yet. There's little reason to risk a working setup to move faster than necessary.

02

For new projects, benchmark Fast mode first

If you're designing something new, line up standard ($10/$50) and Fast ($20/$100) real costs against Sol's current rate, and weigh latency against cost together.

03

Top-tier subscribers: try it in ChatGPT first

Before signing on to the API, check Astra's response quality inside your top-tier ChatGPT plan — it's the cheapest way to inform a later technical migration decision.


06
The Catch

The rollout still isn't even

Because this is a staged rollout, ChatGPT users on lower-tier plans and developers not yet on the API allowlist may not actually be able to try Astra yet, despite the announcement. Right now there's a real gap between "it's rolling out" and "I can use it."

On top of that, OpenAI's pricing has four tiers — standard, cached, batch, and fast — plus a markup for cache writes. Summing that up as a flat "half price" or "double price" oversimplifies it, and your actual bill depends heavily on which modes you combine and how. Whenever the numbers matter, it's worth going back to the primary pricing page to confirm.