GPT-6 Release Date: What Is Actually Confirmed?

Answer from Sep 9See the current answer

Direct links

When was GPT-6 released, and which GPT-6 models can you use today?

Model Name

OpenAI has officially adopted the name GPT-6 Astra (with the API model ID designated as gpt-6-astra) [S1, S2 (Expert C)]. Industry speculation over whether the successor to the GPT-5.6 generation would carry the GPT-6 designation was settled upon launch, confirming that "GPT-6" and "Astra" identify the same core generational model [S3 (Expert C)].

Release Timing and Milestones

GPT-6 Astra has followed a phased rollout approach, distinguishing between announcement and deployment milestones:

Announcement: Formally announced by OpenAI on September 3, 2026 via official release notes and platform updates [S3 (Expert B), S3 (Expert C)].
Gated / Enterprise Availability: Released initially on September 3 as a limited preview to a select group of vetted organizations and enterprise customers via specialized access programs (such as Daybreak) due to rigorous safety evaluations [S4 (Expert C)].
ChatGPT & API Availability: While initial organizational rollouts began immediately, broader availability across ChatGPT paid tiers and developer API access has progressed through staged rollouts over subsequent days and weeks [S3 (Expert A), S5 (Expert C)].

Model Variants & Technical Specifications

Focus and Capabilities: The model is optimized heavily for long-horizon agentic workflows, autonomous computer use, advanced software engineering, and multi-step reasoning tasks [S3 (Expert C)].
Context and Output: It features a 1,050,000-token context window paired with a maximum output length of 128,000 tokens [S3 (Expert C)].
Safety & Alignment Gate: OpenAI designated GPT-6 Astra as its first model to cross the "Critical" cybersecurity threshold under its Preparedness Framework due to autonomous vulnerability-chaining capabilities, leading to strict administrative gating of advanced offensive security tools [S3 (Expert C)].

Benchmarks

Official launch disclosures from OpenAI highlight significant performance gains over predecessor models like GPT-5.6 Sol across autonomous execution, coding, and mathematical reasoning:

Terminal-Bench 4.0: 57.9% [S3 (Expert C)]
OSWorld 2.0 (Computer Use): 72.6% [S3 (Expert C)]
FrontierMath Tier 4: 97.6% – 98% saturation [S3 (Expert C)]
ExploitBench: 100.0% [S3 (Expert C)]
GPQA Diamond: 96.0% [S3 (Expert C)]

(Note: While some early tracking and aggregator summaries referenced higher or unverified estimates, official disclosures establish the benchmarks above [S2 (Expert A), S3 (Expert C)].)

Pricing

Standard API pricing for GPT-6 Astra is structured as follows:

Standard Input: $10 per million tokens [S5 (Expert C)]
Standard Output: $50 per million tokens [S5 (Expert C)]
Caching & Modes: Cached input tokens are priced at $1 per million (with cache writes at $12.50). Batch and Flex processing modes reduce standard costs by half, while high-speed "Fast" mode doubles standard rates [S5 (Expert C)].

Sources

Get an email when this answer changes

Only when a check moves the answer. No account needed.

History

Checked 30 times since Jul 23

How strongly the models agree, per check. Amber points mark checks that changed the answer.

Oct 22
Checked: no change
Read this version
Oct 15
The answer changed

The new answer gives specific rollout timing—an early-September announcement and ChatGPT expansion on October 7–8, 2026—and clarifies which models are surfaced to different ChatGPT tiers. The standing answer described staged availability but gave no dates or tier-specific model access.
Read this version
Sep 30 – Oct 7 3 checks, no change
- Oct 7same answer, new wording Read this version
- Oct 7a different reading, set aside Read this version
- Sep 30 Read this version
Sep 23
The answer changed

The consensus shifted completely from providing specific details about a released model named 'GPT-6 Astra' (including specs, benchmarks, and pricing) to stating that no such model name, release, or pricing can be confirmed from official sources.
Read this version
Sep 9
The answer changed

The consensus changes from GPT-6 being entirely unconfirmed and unreleased to GPT-6 Astra being officially announced, partially deployed, and documented with specific capabilities, benchmarks, context limits, safety classification, and pricing.
Shown above
Sep 2
The answer changed

The new answer omits and effectively contradicts the old answer’s central claim that OpenAI officially confirmed an internal unreleased model called Astra. It instead states that no successor has been officially announced, changing the conclusion about what OpenAI has publicly acknowledged.
Read this version
Jul 24 – Aug 26 21 checks, no change
- Aug 26same answer, new wording Read this version
- Aug 19 Read this version
- Aug 11 Read this version
- Aug 10same answer, new wording Read this version
- Aug 9 Read this version
- Aug 8 Read this version
- Aug 7 Read this version
- Aug 6same answer, new wording Read this version
- Aug 5 Read this version
- Aug 4 Read this version
- Aug 3 Read this version
- Aug 2 Read this version
- Aug 1same answer, new wording Read this version
- Jul 31 Read this version
- Jul 30 Read this version
- Jul 29 Read this version
- Jul 28 Read this version
- Jul 27 Read this version
- Jul 26same answer, new wording Read this version
- Jul 25 Read this version
- Jul 24 Read this version
Jul 23, 2026
Topic started

The first answer. Every later check is compared with it.
Read this version

How the 3 models compared · 5 contradictions

Google Gemini gemini-3.5-flash-lite · DeepSeek deepseek-v4-flash · Grok grok-4.3-no-reasoning

Contradiction · critical

The API model identifier `gpt-6-astra` has been officially designated.

No API model identifier for GPT-6 Astra appears in the documentation examined.

Grok

“No API pricing, rate-card entry, or model identifier for GPT-6 Astra (or any GPT-6 variant) appears”

The API model ID is `gpt-6-astra`.

Gemini

“API model ID: `gpt-6-astra`”

How to verify: Check OpenAI platform documentation for an official `gpt-6-astra` model ID.

Contradiction · critical

GPT-6 Astra has reached developer API availability through a staged rollout.

The available documentation does not verify any API availability.

Grok

“no API availability”

API availability was reportedly provided to existing developers and enterprise accounts at launch, with staged expansion.

DeepSeek

“Reported as available to existing developers/enterprise accounts at launch”

General developer availability began alongside staged ChatGPT rollouts.

Gemini

“alongside general developer availability through the OpenAI API”

How to verify: Confirm whether GPT-6 Astra is actually listed and accessible in OpenAI’s API documentation and console.

Contradiction · critical

OpenAI has published the model’s technical specifications, including its context and output limits.

Official release notes provide no quantitative technical specifications.

Grok

“No benchmark scores, parameter counts, training details, or model-card links are provided”

The 1,050,000-token context and 128,000-token output figures are only aggregator claims and remain unverified.

DeepSeek

The model has a 1,050,000-token context window and 128K maximum output.

Gemini

“Features a **1,050,000-token context window** with a maximum output length of 128K tokens.”

How to verify: Check the official model documentation for the exact context-window and maximum-output values.

Contradiction · critical

Official OpenAI disclosures establish the listed benchmark results.

OpenAI documentation examined publishes no quantitative benchmark results.

Grok

“OpenAI’s documentation does not publish any quantitative performance numbers for GPT-6 Astra.”

The benchmark figures are aggregator claims and should be treated as unverified.

DeepSeek

OpenAI’s official launch disclosures publish the listed benchmark results.

Gemini

“According to OpenAI’s official launch disclosures”

How to verify: Verify each benchmark number against an official OpenAI launch disclosure or model card rather than secondary sources.

Contradiction · critical

The standard API pricing and caching or processing-mode rates are established by OpenAI.

No GPT-6 Astra pricing or API rate-card entry was found.

Grok

“No pricing information, API rate-card entries, or ChatGPT plan-tier pricing for GPT-6 Astra is present”

The $10/$50 pricing and roughly 1.05-million-token context are reported but unverified.

DeepSeek

“Reported API pricing: $10 / $50 per 1M tokens”

The listed input, output, caching, batch, Flex, and Fast rates are established.

Gemini

“Standard API pricing for GPT-6 Astra is established”

How to verify: Check OpenAI’s current API pricing page for the input, output, cache, Batch, Flex, and Fast rates.