GPT-6 Astra API Pricing: $10/$50 per 1M, the 272K Repricing Rule, and a 1.05M-Token Context
GPT-6 Astra price, in one paragraph
GPT-6 Astra API pricing is $10 per 1M input tokens and $50 per 1M output tokens on the standard tier, with cached input at $1 per 1M and cache writes at $12.50 per 1M. The cost twist is the context threshold: prompts with more than 272K input tokens are repriced for the entire request at 2x the input and cache rates and 1.5x the output rate — $20 input / $75 output / $2 cached / $25 cache writes per 1M. GPT-6 Astra carries a 1,050,000-token context window (922K max input, 128K max output), which is exactly why the 272K rule matters: this is OpenAI's first model where a single prompt can cross the threshold many times over. It is now a selectable strategy in the AI Agency Pricing Calculator.
The full GPT-6 Astra rate card (verified Sept 3, 2026)
All figures are USD per 1M tokens from OpenAI's official model page (gpt-6-astra) and pricing page. "Standard" is the default tier; Batch and Flex are priced at 50% of Standard; Fast mode is 2x the applicable rates.
| Tier | Input | Cached input | Cache writes | Output |
|---|---|---|---|---|
| Standard — up to 272K input tokens | $10.00 | $1.00 | $12.50 | $50.00 |
| Standard — over 272K input tokens | $20.00 | $2.00 | $25.00 | $75.00 |
| Batch / Flex — up to 272K | $5.00 | $0.50 | $6.25 | $25.00 |
| Batch / Flex — over 272K | $10.00 | $1.00 | $12.50 | $37.50 |
| Fast mode — up to 272K | $20.00 | $2.00 | $25.00 | $100.00 |
| Fast mode — over 272K | $40.00 | $4.00 | $50.00 | $150.00 |
Reasoning tokens bill at output rates, so a max-effort Astra run can burn output tokens well beyond the visible answer. Cache writes are billed at 1.25x the uncached input rate, so a prefix you cache but reuse only once is more expensive than fresh input — caching pays only after roughly two reads.
The 272K rule: input doubles, output goes 1.5x — for the whole request
OpenAI's model documentation states it precisely: "Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request." Three consequences for agency cost math:
- The threshold is on input tokens, and it reprices everything. Once a prompt exceeds 272K input tokens, the entire request — including the output — bills at the long-context rates, not just the tokens past the threshold.
- Output is 1.5x, not 2x. The "$10/$50 doubles past 272K" shorthand is wrong on output: the verified long-context output rate is $75 per 1M, not $100. Input and cache genuinely double ($10→$20, $1→$2); output rises 50%.
- 1.05M context makes this a real cost cliff, not a footnote. GPT-6 Astra's 922K max input is 3.4x the 272K threshold. Long-document agents, codebase-scale analysis, and research runs that load large contexts can land entirely in the repriced band.
Model details and rollout
GPT-6 Astra (gpt-6-astra, default snapshot gpt-6-astra) is OpenAI's most capable model, built for the hardest end-to-end work — complex reasoning, coding, computer use, research, and document creation. It supports text and image input with text output, reasoning.effort from low through max, a 1,050,000-token context window, a 922,000-token maximum input, and a 128,000-token maximum output (knowledge cutoff Apr 30, 2026). It works through Chat Completions, Responses, and Batch APIs (no Realtime, no Assistants, no fine-tuning yet), with prompt caching, structured outputs, function calling, file search, web search, and computer-use tools.
Availability: GPT-6 Astra began rolling out Sept 3, 2026 to enterprises in OpenAI's Trusted Access Program — including organizations in the Daybreak Access cybersecurity program — with access through ChatGPT Plus, Pro, Business, and Enterprise, the OpenAI API, and AWS arriving in the coming days, per OpenAI's developer documentation and launch coverage. OpenAI calls GPT-6 Astra its first model to reach the Critical level of cybersecurity capability under its Preparedness Framework, and it is positioned as a computer-use flagship — "Anything you can do on a computer, Astra can do for you."
What GPT-6 Astra costs per task (worked examples)
The pricing headline understates the real cost question: most agency tasks stay under 272K input tokens, but GPT-6 Astra's reason to exist is the long, tool-driven runs that don't. Two worked examples on official rates:
| Workload | GPT-6 Astra | GPT-5.6 Sol (promo $4/$20) | Delta |
|---|---|---|---|
| 10K-in / 2K-out task (sub-272K) | ~$0.20 ($0.10 in + $0.10 out) | ~$0.08 ($0.04 in + $0.04 out) | Astra ~2.5x Sol |
| 300K-in / 5K-out run (past 272K → full request repriced) | ~$6.375 ($6.00 in at $20/1M + $0.375 out at $75/1M) | n/a below — Sol long-context input bills $8/1M → ~$2.50 | ~2.5x Sol; ~2x Astra's own sub-272K price for the same tokens |
The second row is the number to internalize: the same 300K-in / 5K-out request priced at sub-272K rates would cost about $3.25 ($3.00 in at $10/1M + $0.25 out at $50/1M). Crossing 272K pushes it to $6.375 ($6.00 in at $20/1M + $0.375 out at $75/1M) — the input line doubles while the output line rises 50%, a ~96% bill increase on the same tokens. For a fleet of long-context agents, that is a cliff driven entirely by the threshold, which is why the calculator models the sub-272K and above-272K bands separately. (Long-context rates for GPT-5.6 Sol per OpenAI's pricing page: $8 input / $30 output per 1M; illustrative math only — verify before quoting.)
What agencies should do with the GPT-6 Astra price
- Quote Astra only where the capability is worth the premium. At 2.5x Sol's promo on both axes, GPT-6 Astra is a top-tier flagship price. Route short, well-scoped tasks to Sol, mid-tier models, or open-weight APIs; save Astra for computer-use, long-horizon research, and hardest-end-to-end work.
- Budget the 272K cliff explicitly. For any workload that can exceed 272K input tokens, compute the repriced band ($20/$75) — not the headline $10/$50. A retainer priced on standard rates will be underquoted by ~90% on that class of request.
- Cache deliberately. At $1/1M cached input vs $12.50/1M cache writes, a cache prefix needs at least two reads to pay. Reused system prompts and shared context are the win; one-shot writes are a loss.
- Watch availability before promising delivery. Access is staged — Trusted Access/Daybreak today, broader API/AWS/consumer rollout "in the coming days." Don't commit client timelines to Astra-only stacks until the tier you need is actually reachable.
- Model it in the calculator. GPT-6 Astra is a selectable Model Strategy on the AI Agency Pricing Calculator — the GPT-6 Astra scenario shows sub-272K vs above-272K cost math using OpenAI's exact repricing rule.
Sources
- OpenAI developer docs — GPT-6 Astra model page (model ID, context window, max input/output, pricing note, endpoints, tools): developers.openai.com/api/docs/models/gpt-6-astra
- OpenAI developer docs — pricing page (Standard/Batch/Flex/Fast tables incl. gpt-6-astra long-context rates): developers.openai.com/api/docs/pricing
- OpenAI — safety overview for GPT-6 Astra (Critical-tier cyber capability, rollout framing): openai.com/index/safety-overview-gpt-6-astra
- OpenAI X post (Sept 3, 2026 launch): x.com/OpenAI
- Coverage used for rollout/positioning context only (not pricing): CNBC (Sept 3, 2026), VentureBeat, 9to5Mac, WIRED
Accuracy note: All token prices, the 272K repricing rule, cache-write rate, context window (1,050,000), max input (922,000), and max output (128,000) verified against OpenAI's official model documentation and pricing page on Sept 3, 2026. Batch/Flex = 50% of Standard and Fast mode = 2x per the same pricing page. Rollout sequence per OpenAI's model docs and launch coverage. No benchmark scores are asserted on this page — pricing only. 2026 API pricing moves quickly; re-verify OpenAI's pricing page before quoting.
Model GPT-6 Astra in your next quote
Try the AI Agency Pricing Calculator →Compare per-task costs across frontier and open-weight models on the AI Model Cost per Task 2026 page.
Frequently asked questions
How much does GPT-6 Astra cost per 1M tokens?
$10 per 1M input / $50 per 1M output / $1 cached input / $12.50 cache writes on the standard tier (≤272K input). Batch and Flex are 50% of standard; Fast mode is 2x.
Does GPT-6 Astra pricing double past 272K tokens?
Input and cache rates double ($10→$20, $1→$2) but output goes 1.5x ($50→$75) — and the whole request is repriced once the 272K input threshold is crossed. The "$10/$50 doubles" shorthand is wrong on output.
What is the GPT-6 Astra context window?
1,050,000 tokens per OpenAI's official model docs (922K max input, 128K max output). Press coverage commonly rounds it to about 1.1M tokens.
When is GPT-6 Astra available?
Rollout began Sept 3, 2026 to Trusted Access / Daybreak organizations. ChatGPT Plus/Pro/Business/Enterprise, the OpenAI API, and AWS follow in the coming days.
How does GPT-6 Astra compare to GPT-5.6 Sol on cost?
At standard rates Astra is 2.5x Sol's promotional $4/$20 on both axes (~$0.20 vs ~$0.08 on a 10K-in/2K-out task). Past 272K input the repricing rule widens the gap — the calculator models both bands.
Where does GPT-6 Astra show up in the pricing calculator?
It is a selectable Model Strategy on the homepage calculator, modeled as the premium OpenAI flagship tier, plus a GPT-6 Astra long-context scenario that applies the 272K repricing rule to sub- and above-272K workloads.