GPT-6 Astra

OpenAI's official launch page confirms that GPT-6 Astra was introduced as "the world's most intelligent and aligned model" and is now available in ChatGPT Work, Codex, and the OpenAI API. The page positions Astra as state-of-the-art across computer use, browsing, professional work, software engineering, cybersecurity, The page also surfaces early enterprise feedback, including a quote from Cognition's Silas Alberti stating that Astra is being integrated into Devin on launch day with state-of-the-art performance on Cognition's internal testing benchmark, citing improved computer use, writing, and codebase understanding. OpenAI notes

This technical deep dive states that OpenAI released GPT-6 Astra on September 3, 2026, with general availability on September 10, 2026, pricing it at $10/$50 per million tokens with a 1.05M-token context window. Architecturally, the article describes Astra as employing a recurrent depth / looped transformer design that The piece reports that Astra saturates multiple benchmarks including ARC-AGI-3 (99.9%), ExploitBench (100%), and FrontierMath Tier 4 (97.6%), and sets new records on computer use (OSWorld 72.6%), coding (Terminal-Bench 4.0: 57.9%), and scientific reasoning (GPQA Diamond: 96.0%). It positions Astra as a new generation —

Coursiv observes that OpenAI does not publish numerical benchmark scores for GPT-6 Astra and instead provides qualitative ratings and precise technical specifications: a 1,050,000-token context window, 128,000 maximum output tokens, a knowledge cutoff of April 30, 2026, text-and-image input with text output, and five r The same piece notes that GPT-5.6 Sol, Terra, and Luna share Astra's 1,050,000-token context and 128,000-output ceiling, framing Astra as the flagship of the broader Astra family rather than a narrow tool. Coursiv is a secondary analysis paraphrasing OpenAI's model documentation, so the specs are best treated as a rest

On September 3, 2026, OpenAI unveiled GPT-6 Astra, describing it as its most intelligent and aligned language model yet, with the model rolling out to ChatGPT Plus, Pro, Business, and Enterprise users and also available through the OpenAI API and AWS. According to the guide, Astra advances beyond previous flagships lik The guide reports OpenAI-published scores including 98% on FrontierMath Tier 4 (v2), 99.9% on ARC-AGI-3, and 100% on ExploitBench, along with major speed gains over GPT-5.6 Sol in computer-use and automation benchmarks (OSWorld 2.0 at 72.6% in about 40 minutes per task vs. 65.7% in about 75 minutes). OpenAI also highli

The Shattered.io write-up, dated September 8, 2026, reports that OpenAI turned on GPT-6 Astra for its first wave of users on September 3, 2026, and frames the release as a flagship entry in a broader Astra model family referenced internally as Sol, Terra, and Luna. It quotes OpenAI's launch announcement on staged rollo The piece notes that Astra is the first OpenAI model to reach the "Critical" tier of cybersecurity capability under the company's Preparedness Framework and describes the gated access structure as unusual for a mainstream product launch. According to the excerpt, Astra is being pushed into ChatGPT Plus, Pro, Business,

Founderz AI Business School provides a technical access and pricing breakdown for GPT-6 Astra, which it calls OpenAI's flagship model for complex, end-to-end work, launched September 3, 2026. The developer model identifier is gpt-6-astra, priced at USD 10 per 1M input tokens, USD 1 per 1M cached input, and USD 50 per 1 The piece clarifies the distinction between GPT-6 Astra (the model) and ChatGPT (the product), noting that Astra access depends on plan tier: Free/Go has no documented Astra access, Plus includes Astra in Work and Codex, eligible Pro plans can use GPT-6 Pro in Chat, and Business and Enterprise access depends on seat an

A Medium analysis by unicodeveloper summarizes GPT-6 Astra as OpenAI's new frontier model released September 3, 2026, describing it as state-of-the-art in computer use, browsing, software engineering, cybersecurity, science, professional work, and 3D/CAD-style tasks. The piece highlights Astra's computer-operator frami The post reproduces OpenAI's published benchmark numbers, including 95.9% on BenchCAD, 97.6% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on ExploitBench, while noting that Claude Fable 5.1 beats Astra on Humanity's Last Exam (with tools) and on the Artificial Analysis Intelligence Index in OpenAI's own compari

A September 5, 2026 Medium analysis by Mehul Gupta recaps OpenAI's GPT-6 Astra launch and the headline benchmark figures that dominated coverage: 99.9% on ARC-AGI-3, 97.6% on FrontierMath Tier 4, and 100% on ExploitBench. The piece notes that of the fourteen benchmarks OpenAI highlighted in its announcement, only one o The analysis argues for looking past composite scores to implementation details when evaluating model progress, positioning itself as a case study rather than a takedown of Astra. It acknowledges that GPT-6 Astra represents a genuine step forward in some respects, while encouraging developers and evaluators to scrutini

Vellum's benchmark explainer confirms OpenAI released GPT-6 Astra on September 3, 2026, with Greg Brockman telling reporters it is "not unreasonable to feel that we are now in the AGI era." The launch was OpenAI's largest training run to date, run on over 100,000 GPUs at the Stargate site in Texas, and marked the first On computer-use benchmarks, Vellum reports GPT-6 Astra scoring 59.3% on Agents' Last Exam (vs. 53.6 for GPT-5.6 Sol), 72.6% on OSWorld 2.0 (vs. 65.7% for Sol, with task time cut roughly 47% to about 40 minutes), and 92.7% on ScreenSpot-Pro (vs. 76.9% for Sol). The piece also notes that not every table goes Astra's way:

LLM-Stats tracks GPT-6 Astra with a composite LLM Stats Score of 59.6 at a blended price of $11.9 per million tokens, and includes a quality-tracker sigma curve showing -0.16σ over 7 days, -0.23σ over 30 days, and +15.91σ versus baseline, sourced from 43 votes. The page lists GPT-6 Astra at the high end of the cost-eff The aggregator also tracks performance by conversation depth and places GPT-6 Astra within capability tiers, but the composite page does not cite OpenAI's primary blog post directly and contains no Opper-specific content. Figures such as the $11.9/M blended price reflect the aggregator's own composite, not OpenAI list

MindStudio's benchmark comparison reports GPT-6 Astra scoring approximately 99% on ARC-AGI-3 (up from 7.8% on GPT-5.6, compared to a 48% average human score), 97.6% on FrontierMath Tier 4 versus 87.8% for Claude Fable 5.1, 100% on ExploitBench, and roughly 73–74% on DeepSWE. On DeepSWE, Gemini 3 Flash (73.7–74%) and Me MindStudio also documents Astra's API pricing at $10 per million input tokens and $50 per million output tokens, with a "fast mode" offering 2.5x speed for 2x price and early-tester estimates of roughly $167 per aggregated task. Alignment testing reportedly showed 0% unauthorized task completion versus 48.2% for the pr

MindStudio's independent analysis finds that GPT-6 Astra is "not consistently" better than Claude Fable 5.1, beating it on computer use, terminal workflows, and long-context retrieval while trailing on broader indices. On Artificial Analysis's Intelligence Index, Fable 5.1 scores 66 versus Astra's 61, and on the Coding The critique surfaces a methodology issue: Astra's headline ARC-AGI-3 score of 99.9% uses OpenAI's provider-adapter harness, which preserves reasoning state between actions, whereas a standard neutral harness puts Astra at 62.7%. Astra costs about 2.5x more per token than GPT-5.6 Sol ($10/$50 vs $4/$20), and even with