What is Astra Ultrafast list price?

$60 input, $6 cached input, $75 cache write, and $300 output per million tokens. That is 6× Astra Standard ($10 / $1 / $12.50 / $50). Retrieved 2026-10-02 from the OpenAI pricing Ultrafast tab.

How do I call Astra Ultrafast?

Set model to gpt-6-astra and service_tier to ultrafast. It is a serving tier on the existing model, not a new model ID. WebSockets are recommended for tool-heavy agent loops.

What are the default Ultrafast TPM limits?

500,000 tokens per minute for API usage tiers 1 to 3, 1,000,000 for tier 4, and 5,000,000 for tier 5. Higher limits go through your OpenAI account team.

Does Ultrafast support EU data residency?

No. Ultrafast supports US data residency and global processing only. There are no EU or other non-US regional endpoints.

Is API Ultrafast the same as ChatGPT Pro 500 Ultrafast?

No. Pro 500 Ultrafast is ChatGPT seat packaging. API Ultrafast is billed per token at the service_tier rates above. Keep them on separate lines.

Is Sol Ultrafast priced?

Yes. Sol Ultrafast is now live as of Oct 8 2026. See [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form).

← Reality Check

GPT-6 Astra Ultrafast: 6x Price and Order Form

OpenAI docs (retrieved 2026-10-02) + NVIDIA Oct 1 2026: GPT-6 Astra Ultrafast is broadly available in the API. Short-context list is $60 input / $6 cached / $75 cache write / $300 output per MTok (6× Astra Standard $10/$1/$12.50/$50). Set model gpt-6-astra and service_tier ultrafast. Default Ultrafast TPM: 500k (tiers 1–3), 1M (tier 4), 5M (tier 5). US residency + global only; no EU / non-US regional. Vendor claims up to 8× Standard generation speed. Map tier, rates, TPM, region, chargeback, and accepted-outcome owner before Ultrafast is the agent default. Soft subscribe only.

Jonathan Kvarfordt · Published October 2, 2026 · Last reviewed October 8, 2026 · 8 min read

Why trust this analysis?

The short answer

When Astra Ultrafast is 6× Standard list, which Order Form lines must we name before agents default to service_tier ultrafast?

Lock gpt-6-astra + service_tier ultrafast, short list $60/$6/$75/$300, default TPM by usage tier, US/global-only region gate, WebSocket transport for tool loops, chargeback owner, and accepted-outcome cost. Do not mash with Pro 500 seat Ultrafast and keep Sol rates on [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form).

Decision rule

Budget the 6× list and the TPM and region gates, not the tok/s headline. Name service_tier, rates, TPM, region, chargeback, and accepted outcome before Ultrafast is the default route.

Evidence

developers.openai.com/api/docs/pricing (Ultrafast tab, retrieved 2026-10-02): gpt-6-astra short-context Ultrafast $60 input / $6 cached / $75 cache write / $300 output per MTok; Standard $10 / $1 / $12.50 / $50; Fast $20 / $2 / $25 / $100
developers.openai.com/api/docs/guides/ultrafast-mode (retrieved 2026-10-02): broadly available for GPT-6 Astra at low default rate limits; preview for GPT-5.6 Sol; default TPM 500,000 (tiers 1–3), 1,000,000 (tier 4), 5,000,000 (tier 5); US data residency and global processing only; WebSockets recommended for tool loops
blogs.nvidia.com (Oct 1, 2026): Astra Ultrafast available in the OpenAI API; up to 8× Standard token generation speed. VENDOR claim

Operator action

Today: write gpt-6-astra + service_tier ultrafast on the Order Form, lock the short list rates, record your TPM tier, check the region gate, name the chargeback owner, and measure cost per accepted agent task before Ultrafast becomes the default.

Last reviewed October 8, 2026

Six times the sticker is not a feature. It is a bill line.

GPT-6 Astra Ultrafast is now broadly available in the OpenAI API. Someone on your team will flip an agent to it because it feels fast. This decoder covers what that flip costs and which lines belong on paper first.

The argument

How this reality check breaks down

A map of the sections ahead, in the order the case is made. The sourced charts are in the research library.

What landed

NVIDIA said on Oct 1 2026 that Astra Ultrafast is available in the OpenAI API and for eligible ChatGPT Work and Codex users (NVIDIA). The OpenAI Ultrafast mode guide (retrieved 2026-10-02) says it is broadly available for GPT-6 Astra at low default rate limits, with preview access only for GPT-5.6 Sol (OpenAI Ultrafast guide).

You call it with model `gpt-6-astra` and `service_tier: ultrafast`. It is a serving tier, not a new model ID.

Reality Check

GPT-6 Astra Ultrafast: 6x Price and Order Form

Source-cited charts live in the research library.

The list ladder (locked)

Short-context list, $ per million tokens (MTok), from the pricing page Ultrafast tab, retrieved 2026-10-02 (OpenAI pricing). Order: input / cached input / cache write / output.

Standard: $10 / $1 / $12.50 / $50 (baseline).
Fast: $20 / $2 / $25 / $100 (2× Standard).
Ultrafast: $60 / $6 / $75 / $300 (6× Standard).
Batch / Flex (context only): $5 / $0.50 / $6.25 / $25.

Astra API serving tiers: Standard $10 input, $1 cached, $12.5 cache write, $50 output per MTok; Fast 2× at $20, $2, $25, $100; Ultrafast 6× at $60, $6, $75, $300. — gpt-6-astra short-context list per MTok. Retrieved 2026-10-02 from OpenAI pricing. Speed claim is vendor.

Long context is a soft open. Astra has an over-272K input band in general. Ultrafast long-context rates were not verified for this piece. Confirm them on the pricing Ultrafast tab before you print them on an Order Form.

Throughput ceiling

Default Ultrafast tokens per minute (TPM) by API usage tier, per the guide:

Tiers 1–3: 500,000 TPM.
Tier 4: 1,000,000 TPM.
Tier 5: 5,000,000 TPM.
Higher limits or Sol preview: request through your OpenAI account team.

A volume agent that hits the ceiling falls back or queues. Know which happens before go-live.

Region gate

Ultrafast supports US data residency and global processing only. No EU or other non-US regional endpoints. If your contract requires EU residency, Ultrafast is off the table. That is a hard deployment gate, not a pricing question.

Transport

OpenAI strongly recommends WebSockets for agentic tool loops. HTTP works, but connection overhead on every tool call can erase the latency you paid 6× to get.

The speed claim

OpenAI and NVIDIA claim up to 8× Standard token generation speed. Tag it VENDOR. It measures tokens out per second, not end-to-end task time, and it is not a guarantee. Your agent spends time in tool calls, retrieval, and retries too.

Decision rule

Budget the 6× list and the TPM and region gates, not the tok/s headline. Name service_tier, rates, TPM, region, chargeback, and accepted outcome before Ultrafast is the default route. Cost per accepted agent task is the number that matters. See the dictionary for terms.

Order Form checklist

Model + tier: gpt-6-astra + service_tier ultrafast (not a new model ID).
List rates: short Ultrafast $60 / $6 / $75 / $300 per MTok.
TPM ceiling: default 500k / 1M / 5M by API usage tier; raises via account team.
Region gate: US residency + global only. No EU or non-US regional endpoints.
Transport: WebSockets for tool-heavy loops; HTTP works.
Chargeback owner: who owns Ultrafast route spend versus Standard and Fast.
Accepted outcome: cost per accepted agent task, not tok/s alone.
Distinct meters: not Pro 500 seat Ultrafast. Sol Ultrafast is now live as of Oct 8 2026; see Sol Ultrafast Order Form decoder.

Keep the meters apart

ChatGPT Pro seat Ultrafast is seat packaging. See Pro 200 usage cut and Pro 500 Ultrafast.
GPT-6.1 Sol is a separate list at one-fifth of Astra. Sol Ultrafast is now live as of Oct 8 2026; see Sol Ultrafast Order Form decoder. See Sol near Astra at one-fifth price.
This piece covers Astra API Ultrafast serving tier rates, TPM, and region only.

If you want the next Order Form decoder before agents burn the 6× tier, subscribe.

Take it to the room

The short list from this piece

Pulled from the argument above, written so you can read it out in a pipeline or board review.

Frequently asked questions

What is Astra Ultrafast list price?: $60 input, $6 cached input, $75 cache write, and $300 output per million tokens. That is 6× Astra Standard ($10 / $1 / $12.50 / $50). Retrieved 2026-10-02 from the OpenAI pricing Ultrafast tab.
How do I call Astra Ultrafast?: Set model to gpt-6-astra and service_tier to ultrafast. It is a serving tier on the existing model, not a new model ID. WebSockets are recommended for tool-heavy agent loops.
What are the default Ultrafast TPM limits?: 500,000 tokens per minute for API usage tiers 1 to 3, 1,000,000 for tier 4, and 5,000,000 for tier 5. Higher limits go through your OpenAI account team.
Does Ultrafast support EU data residency?: No. Ultrafast supports US data residency and global processing only. There are no EU or other non-US regional endpoints.
Is API Ultrafast the same as ChatGPT Pro 500 Ultrafast?: No. Pro 500 Ultrafast is ChatGPT seat packaging. API Ultrafast is billed per token at the service_tier rates above. Keep them on separate lines.
Is Sol Ultrafast priced?: Yes. Sol Ultrafast is now live as of Oct 8 2026. See [Sol Ultrafast Order Form decoder](/blog/gpt-6-1-sol-ultrafast-6x-api-order-form).

Share this page

Posting to Instagram or TikTok? Copy the link, it carries the title, summary and share image.

Get the next Reality Check before you sign the order form.

Keep reading

Reality Check

Three Tests Before Your Next AI Purchase: Result, Trust, Ownership

Before you buy AI, answer three questions: what result improves, how much checking remains, and who owns it. Proof rules and Order Form lines inside.

Reality Check

AI Review Time Is a Line Item. Price It and Name the Owner.

On TextQL's Argo-Bench the best AI model fully solved 34.8% of 210 data tasks. Put review time and a named owner into your AI business case.