5f:I[74892,["/_next/static/immutable/chunks/3vxhblv14tjpf.js","/_next/static/immutable/chunks/1a4tp1bvkv9bm.js","/_next/static/immutable/chunks/2_h5l5miz7e61.js","/_next/static/immutable/chunks/12eh5aect5sq2.js","/_next/static/immutable/chunks/3-tsmo25icv4a.js","/_next/static/immutable/chunks/0de-m41gt6dml.js","/_next/static/immutable/chunks/14pvwbbyqzisq.js","/_next/static/immutable/chunks/3x6v819vh-0cy.js","/_next/static/immutable/chunks/3u8_llycdnijy.js","/_next/static/immutable/chunks/0a5gsof_h_wz-.js","/_next/static/immutable/chunks/0d2ui1kf_nby3.js","/_next/static/immutable/chunks/3q9y0fvfy7sys.js"],"ShareButtons"] 60:I[7025,["/_next/static/immutable/chunks/3vxhblv14tjpf.js","/_next/static/immutable/chunks/1a4tp1bvkv9bm.js","/_next/static/immutable/chunks/2_h5l5miz7e61.js","/_next/static/immutable/chunks/12eh5aect5sq2.js","/_next/static/immutable/chunks/3-tsmo25icv4a.js","/_next/static/immutable/chunks/0de-m41gt6dml.js","/_next/static/immutable/chunks/14pvwbbyqzisq.js","/_next/static/immutable/chunks/3x6v819vh-0cy.js","/_next/static/immutable/chunks/3u8_llycdnijy.js","/_next/static/immutable/chunks/0a5gsof_h_wz-.js","/_next/static/immutable/chunks/0d2ui1kf_nby3.js","/_next/static/immutable/chunks/3q9y0fvfy7sys.js"],"CopyArticleButton"] 64:I[93127,["/_next/static/immutable/chunks/3vxhblv14tjpf.js","/_next/static/immutable/chunks/1a4tp1bvkv9bm.js","/_next/static/immutable/chunks/2_h5l5miz7e61.js","/_next/static/immutable/chunks/12eh5aect5sq2.js","/_next/static/immutable/chunks/3-tsmo25icv4a.js","/_next/static/immutable/chunks/0de-m41gt6dml.js","/_next/static/immutable/chunks/14pvwbbyqzisq.js","/_next/static/immutable/chunks/3x6v819vh-0cy.js","/_next/static/immutable/chunks/3u8_llycdnijy.js","/_next/static/immutable/chunks/0a5gsof_h_wz-.js","/_next/static/immutable/chunks/0d2ui1kf_nby3.js","/_next/static/immutable/chunks/3q9y0fvfy7sys.js"],"TableOfContents"] 65:I[97680,["/_next/static/immutable/chunks/3vxhblv14tjpf.js","/_next/static/immutable/chunks/1a4tp1bvkv9bm.js","/_next/static/immutable/chunks/2_h5l5miz7e61.js","/_next/static/immutable/chunks/12eh5aect5sq2.js","/_next/static/immutable/chunks/3-tsmo25icv4a.js","/_next/static/immutable/chunks/0de-m41gt6dml.js","/_next/static/immutable/chunks/14pvwbbyqzisq.js","/_next/static/immutable/chunks/3x6v819vh-0cy.js","/_next/static/immutable/chunks/3u8_llycdnijy.js","/_next/static/immutable/chunks/0a5gsof_h_wz-.js","/_next/static/immutable/chunks/0d2ui1kf_nby3.js","/_next/static/immutable/chunks/3q9y0fvfy7sys.js"],"ArticleContent"] 53:["$","span",null,{"className":"font-mono font-normal text-faint","children":["· ",7," ","min read"]}] 54:["$","h1",null,{"children":"GPT-6 Astra Is Here: What Changed and How to Use It"}] 55:["$","p",null,{"className":"standfirst","children":"OpenAI's GPT-6 Astra focuses on computer use, coding, research, and long workflows. See who gets access, what it costs, and how to prompt it well today."}] 56:["$","div",null,{"className":"art-byline","children":[["$","span",null,{"className":"mono-av","aria-hidden":"true","children":"A"}],["$","b",null,{"children":"By Assaf Biton"}],[["$","span",null,{"className":"dot","aria-hidden":"true"}],["$","time",null,{"className":"font-mono","children":"September 7, 2026"}]],false]}] 61:T2d18,

GPT-6 Astra is not just a new answer engine. OpenAI built its September 2026 flagship for work that continues across browsers, code, documents, and professional software. The useful question is no longer only “Is it smarter?” It is “Which job should I trust it to move forward, and how do I keep control?”

Editorial illustration of GPT-6 Astra coordinating work across code, documents, research and browser tools — GPT-6 Astra was announced on September 3, 2026. AI Deck editorial illustration.

What matters

Astra is designed for long, multi-step work and can accept direction while a task is already running. Choose it when the job crosses tools, needs careful verification, or must preserve a large amount of context. For a quick answer or a simple rewrite, a faster model will usually be the more sensible choice.

What OpenAI actually launched

OpenAI announced GPT-6 Astra on September 3, 2026. Access started with a limited group of organizations and is rolling out to ChatGPT Plus, Pro, Business, and Enterprise users, as well as the OpenAI API, Microsoft Azure, and AWS Bedrock. Enterprise administrators must enable Astra because it is off by default at launch.

The API model ID is gpt-6-astra. It supports a 1,050,000-token context window and up to 128,000 output tokens. Standard API pricing is $10 per million input tokens and $50 per million output tokens, before tool-call costs and long-context surcharges.

Change	What it means in practice	Use it when
Computer use	The model can work directly in browsers and professional software instead of only describing the next click.	The task crosses several apps and ends in a result you can review.
Mid-turn steering	You can add a requirement or correction while work continues.	A long task changes after the model has already started.
Async tool calls	The model can continue independent work while your application waits for a slow tool.	Your workflow includes searches, jobs, or external systems with uneven response times.
Large context	More source material, code, and prior work can stay available in one task.	The answer depends on many files or a long-running investigation.
Professional artifacts	OpenAI trained Astra to follow templates for documents, spreadsheets, presentations, and visual work.	You need a usable artifact, not a paragraph describing one.

The benchmark story is strong, but not simple

OpenAI reports large gains on work that looks agentic. Astra scored 57.9% on Terminal-Bench 4.0, compared with 37.3% for GPT-5.6 Sol, and 41.4% on AutomationBench, compared with 18.1% for Sol. On OSWorld 2.0, which evaluates computer interaction, OpenAI reports 72.6% for Astra.

But “best model” is too broad. In OpenAI's own comparison table, Astra scored 61.2 on the Artificial Analysis Intelligence Index. Claude Fable 5.1 scored 65.7 and Claude Opus 5 scored 63.1 in the same table. Astra also trailed some Claude models on the Artificial Analysis Coding Agent Index.

Read benchmarks as a map, not a verdict

The launch evidence points most clearly to computer use, long workflows, science, and tool coordination. It does not prove that Astra will write every email, solve every coding task, or fit every budget better than a cheaper model.

The previous generation is the useful baseline for judging Astra's gains. For another view of long-running agents, the OpenClaw guide explains why task boundaries matter as much as model intelligence.

How to brief GPT-6 Astra for a serious task

Astra's model guide says it is more likely to ask a focused question when missing information could change the result. That is useful for a major decision. It can also create friction if the assignment is vague.

The solution is not a giant prompt. Give it a clear outcome, the material it may use, the boundaries it must respect, the proof you expect, and what it should do when a detail is missing.

A copy-ready Astra task brief

Outcome: Produce [the exact artifact or decision I need].

Context: Use [files, links, prior decisions, and audience].

Quality bar: The result must satisfy [three concrete checks].

Boundaries: Do not publish, purchase, delete, contact anyone, or change production systems.

Autonomy: Make reasonable assumptions for reversible details. Ask only when the answer changes cost, scope, security, or the final outcome.

Verification: Test the result, show the evidence, and state what remains unverified.

Finish: Continue until the checks pass or a real blocker remains.

This structure is especially useful in Codex. Before a large code task, point Astra to the repository instructions, define the files or behavior in scope, and name the tests that prove completion. Our guide to agent teams, skills, and hooks provides a useful comparison for designing that control layer.

When Astra is worth choosing

Good fit	Probably unnecessary	Needs extra care
Repository-wide engineering work with tests	A short factual answer	Any task with production write access
Research across many current sources	A simple rewrite or summary	Financial, legal, health, or security decisions
Documents that must follow a house template	High-volume low-cost classification	Browser work exposed to untrusted page instructions
Long tasks that may need steering midway	A one-step tool call	Work that sends, deletes, pays, or publishes

For ordinary ChatGPT use, start with a task that would normally require several tabs or several rounds of cleanup. Compare the final result and the time you spent reviewing it against your usual model. A larger context window and higher benchmark score only matter if they reduce your total work.

If your main need is writing quality rather than tool use, compare the result with a dedicated writing pass. Our Claude Opus review and Gemini 3.1 Pro review show why model choice still depends on the actual job.

The safety tradeoff is part of the product

Astra is the first OpenAI model classified at the Critical cybersecurity capability threshold. OpenAI says the production model has stronger resistance to prompt injection and fewer severe misalignment flags than GPT-5.6 Sol. At the same time, its safety report describes weaker visibility into parts of the model's written reasoning during adversarial tests.

For users, the immediate consequence is simpler: some tasks may pause or stop for review. That can feel annoying during legitimate work, but it is not random decoration. OpenAI is giving a more capable model access to browsers, code, and external tools, so consequential actions need a visible control point.

Keep the approval where the consequence is

Let Astra research, draft, compare, test, and prepare. Keep a human decision before payments, deletion, messages, production changes, or publication. You do not need approval for every paragraph. You need it at the point where a mistake becomes expensive or public.

What should you do today?

Our recommendation

If Astra appears in your ChatGPT or Codex account, do not begin with a casual chat. Give it one difficult task that crosses sources or tools, use the task brief above, and measure how much review work remains.

If you mainly need quick answers, wait for broader access or use a faster model. If you build with the API, test Astra on a fixed evaluation set before accepting the higher token price. The useful signal is not whether one demo looks impressive. It is whether Astra produces a correct, reviewable result with fewer corrections than your current workflow.

You can follow new model coverage in our AI news section and check ChatGPT's tool page as availability changes.

Frequently asked questions

Is GPT-6 Astra available to everyone?

No. As of September 5, 2026, OpenAI describes a staged rollout. Limited organizations received access first, with Plus, Pro, Business, Enterprise, API, Azure, and AWS Bedrock availability planned over the following days.

How much does GPT-6 Astra cost?

Standard API pricing is $10 per million input tokens and $50 per million output tokens. ChatGPT usage is included within eligible subscription allowances, with additional credits available for some users and businesses.

Is Astra always better than GPT-5.6 Sol or Claude?

No. OpenAI reports strong gains in computer use, automation, science, and several coding tasks, but its own comparison table shows Astra behind some Claude models on broader intelligence and coding-agent indices.

What is the best first task for Astra?

Choose a complex but reversible task: research a current topic, inspect a large repository, or build a document from a real template. Define the output and verification checks, then keep approval before any public or destructive action.

Sources

57:["$","div",null,{"className":"mt-6 flex flex-wrap items-center justify-center gap-4","children":[["$","$L5f",null,{"url":"https://aideck.website/articles/gpt-6-astra-what-changed","title":"GPT-6 Astra Is Here: What Changed and How to Use It","description":"OpenAI's GPT-6 Astra focuses on computer use, coding, research, and long workflows. See who gets access, what it costs, and how to prompt it well today."}],["$","$L60",null,{"title":"GPT-6 Astra Is Here: What Changed and How to Use It","excerpt":"OpenAI's GPT-6 Astra focuses on computer use, coding, research, and long workflows. See who gets access, what it costs, and how to prompt it well today.","content":"$61","category":"AI News","publishedAt":"2026-09-07T12:50:21.871103+00:00","sourceUrl":"https://openai.com/index/gpt-6-astra/","articleUrl":"https://aideck.website/articles/gpt-6-astra-what-changed"}],"$L62","$L63"]}] 58:["$","$L64",null,{"headings":[{"id":"what-matters","text":"What matters","level":2},{"id":"what-openai-actually-launched","text":"What OpenAI actually launched","level":2},{"id":"the-benchmark-story-is-strong-but-not-simple","text":"The benchmark story is strong, but not simple","level":2},{"id":"read-benchmarks-as-a-map-not-a-verdict","text":"Read benchmarks as a map, not a verdict","level":3},{"id":"how-to-brief-gpt-6-astra-for-a-serious-task","text":"How to brief GPT-6 Astra for a serious task","level":2},{"id":"a-copy-ready-astra-task-brief","text":"A copy-ready Astra task brief","level":3},{"id":"when-astra-is-worth-choosing","text":"When Astra is worth choosing","level":2},{"id":"the-safety-tradeoff-is-part-of-the-product","text":"The safety tradeoff is part of the product","level":2},{"id":"keep-the-approval-where-the-consequence-is","text":"Keep the approval where the consequence is","level":3},{"id":"what-should-you-do-today","text":"What should you do today?","level":2},{"id":"our-recommendation","text":"Our recommendation","level":3},{"id":"frequently-asked-questions","text":"Frequently asked questions","level":2},{"id":"is-gpt-6-astra-available-to-everyone","text":"Is GPT-6 Astra available to everyone?","level":3},{"id":"how-much-does-gpt-6-astra-cost","text":"How much does GPT-6 Astra cost?","level":3},{"id":"is-astra-always-better-than-gpt-56-sol-or-claude","text":"Is Astra always better than GPT-5.6 Sol or Claude?","level":3},{"id":"what-is-the-best-first-task-for-astra","text":"What is the best first task for Astra?","level":3},{"id":"sources","text":"Sources","level":2}],"minHeadings":3}] 59:["$","figure",null,{"className":"art-lead","children":["$","div",null,{"className":"art-media","children":["$","$L44",null,{"src":"/images/articles/gpt-6-astra-what-changed-thumbnail.jpg","alt":"GPT-6 Astra Is Here: What Changed and How to Use It","fill":true,"sizes":"(max-width: 1024px) 100vw, 1024px","className":"object-cover","priority":true}]}]}] 66:T2d18,

What matters

What OpenAI actually launched

Change	What it means in practice	Use it when
Computer use	The model can work directly in browsers and professional software instead of only describing the next click.	The task crosses several apps and ends in a result you can review.
Mid-turn steering	You can add a requirement or correction while work continues.	A long task changes after the model has already started.
Async tool calls	The model can continue independent work while your application waits for a slow tool.	Your workflow includes searches, jobs, or external systems with uneven response times.
Large context	More source material, code, and prior work can stay available in one task.	The answer depends on many files or a long-running investigation.
Professional artifacts	OpenAI trained Astra to follow templates for documents, spreadsheets, presentations, and visual work.	You need a usable artifact, not a paragraph describing one.

The benchmark story is strong, but not simple

Read benchmarks as a map, not a verdict

How to brief GPT-6 Astra for a serious task

The solution is not a giant prompt. Give it a clear outcome, the material it may use, the boundaries it must respect, the proof you expect, and what it should do when a detail is missing.

A copy-ready Astra task brief

Outcome: Produce [the exact artifact or decision I need].

Context: Use [files, links, prior decisions, and audience].

Quality bar: The result must satisfy [three concrete checks].

Boundaries: Do not publish, purchase, delete, contact anyone, or change production systems.

Autonomy: Make reasonable assumptions for reversible details. Ask only when the answer changes cost, scope, security, or the final outcome.

Verification: Test the result, show the evidence, and state what remains unverified.

Finish: Continue until the checks pass or a real blocker remains.

When Astra is worth choosing

Good fit	Probably unnecessary	Needs extra care
Repository-wide engineering work with tests	A short factual answer	Any task with production write access
Research across many current sources	A simple rewrite or summary	Financial, legal, health, or security decisions
Documents that must follow a house template	High-volume low-cost classification	Browser work exposed to untrusted page instructions
Long tasks that may need steering midway	A one-step tool call	Work that sends, deletes, pays, or publishes

The safety tradeoff is part of the product

Keep the approval where the consequence is

What should you do today?

Our recommendation

You can follow new model coverage in our AI news section and check ChatGPT's tool page as availability changes.

Frequently asked questions

Is GPT-6 Astra available to everyone?

How much does GPT-6 Astra cost?

Is Astra always better than GPT-5.6 Sol or Claude?

What is the best first task for Astra?

Sources

5a:["$","$L65",null,{"content":"$66","tools":[{"id":"85441a68-5e8f-419e-9846-9c517f55af13","tips":[],"tool":{"id":"67010260-a0fe-40f7-9e82-17ca72a69f99","faq":[{"answer":"ChatGPT offers a free tier with limited features.","question":"Is ChatGPT free?"},{"answer":"ChatGPT is best for general users, developers, businesses. Key use cases include: Writing & editing, Code generation, Research assistance, Customer support, Data analysis.","question":"What is ChatGPT best for?"},{"answer":"ChatGPT's key features include: GPT-4o model, Image generation (DALL-E 3), Code interpreter, Web browsing, Custom GPTs, API access. These features make it a strong choice in its category.","question":"What are the main features of ChatGPT?"},{"answer":"The top alternatives to ChatGPT include: Claude, Gemini. Each offers different strengths depending on your specific needs and budget.","question":"What are the best alternatives to ChatGPT?"},{"answer":"Pros: Largest user community, Most integrations, Best plugin ecosystem, Reliable uptime. Cons: Free tier rate limits, No real-time data on free plan, Can hallucinate facts.","question":"What are the pros and cons of ChatGPT?"}],"url":"https://chat.openai.com","cons":["Plus plan ($20/mo) imposes daily message limits on GPT-5.4; heavy users hit caps before noon","Computer-use feature is US-only and requires desktop app install as of March 2026","Context window degrades in quality above 600K tokens — 1M is theoretical max, not practical","Free tier locks users to older GPT-4o mini model, not GPT-5.4","Codex App autonomous tasks can consume credits unpredictably, reaching $50+ in a single session"],"name":"ChatGPT","pros":["Largest installed base means most third-party integrations and community resources","GPT-5.4 scores top-3 on most public benchmarks as of Q1 2026","Computer-use reduces need for separate automation tools for desktop workflows","Custom GPTs allow non-developers to build specialized assistants without code"],"slug":"chatgpt","tips":["Use Projects to store persistent instructions and reference files so you don't paste context on every session","Switch to GPT-4o mini for simple formatting tasks and save GPT-5.4 credits for complex reasoning","For autonomous coding with Codex App, set a spending cap in settings before running long tasks","Upload structured CSV data directly and ask for Python analysis — ChatGPT will write and execute the code"],"rating":4.5,"name_he":null,"pricing":"freemium","best_for":["General knowledge workers needing a capable all-in-one assistant","Developers wanting quick code generation with GPT Image and Codex integration","Teams already using OpenAI APIs who want consistent model access","Non-technical users who need to build custom workflows without coding"],"category":"AI Assistants","features":["GPT-5.4 model with 1M token context window","Computer-use feature for desktop task automation","Codex App integration for autonomous coding agents","Voice mode with real-time conversation","DALL-E image generation built in as GPT Image","File upload and analysis (PDF, CSV, images, code)","Custom GPTs marketplace with 3M+ published agents","Projects feature for persistent context across sessions"],"logo_url":"https://icons.duckduckgo.com/ip3/openai.com.ico","seo_title":"ChatGPT Review 2026: GPT-5.4 Features & Pricing","use_cases":["Software developer drafting architecture docs and code reviews with 50K+ token context for full codebase","Marketing manager generating and A/B testing ad copy variations across 10+ formats in one session","Data analyst uploading monthly CSV exports and getting automated EDA with charts and recommendations","Student writing research papers with citation checking and multi-document summarization","Operations manager using computer-use to automate repetitive data entry across legacy desktop apps"],"created_at":"2026-02-24T15:03:54.334626+00:00","updated_at":"2026-06-07T21:50:19.009+00:00","category_id":"7b6ae2bc-8552-4395-8215-570aff330f54","description":"OpenAI's general-purpose AI assistant now powered by GPT-5.4 with 1M token context and computer-use capability for desktop automation. Supports text, voice, image, file, and code workflows in one interface. Used by 400M+ weekly active users across professional and consumer contexts.","is_featured":true,"seo_title_he":"סקירת ChatGPT 2026: יכולות ומחירים","description_he":"העוזר הדיגיטלי של OpenAI, מבוסס כעת על GPT-5.4 עם חלון הקשר של מיליון טוקנים ויכולת שליטה במחשב לאוטומציה שולחנית. תומך בטקסט, קול, תמונה, קבצים וקוד בממשק אחד. בשימוש של למעלה מ-400 מיליון משתמשים פעילים שבועיים.","pricing_details":{"pro":"Plus: $20/mo","notes":"Also: Pro: $100/mo (5x) or $200/mo (20x) · Business: $20/user/mo annual ($25 monthly) — Pro split into $100 (5x) and $200 (20x) tiers in April 2026. Business price dropped to $20/user/mo (annual). Enterprise custom.","starter":"Go: $8/mo","free_tier":"Free: $0/mo, GPT-5.3 Instant with usage caps (~10 messages/5 hrs)."},"seo_description":"ChatGPT now runs GPT-5.4 with 1M context and computer-use. Honest review of features, pricing tiers, and real limitations for professionals.","learning_resources":null,"seo_description_he":"ChatGPT מבית OpenAI עם GPT-5.4, חלון הקשר של מיליון טוקנים ושליטה במחשב. סקירה מקיפה של יכולות, מחירים ומגבלות לאנשי מקצוע."},"tool_id":"67010260-a0fe-40f7-9e82-17ca72a69f99","article_id":"2898042e-f5e7-432b-ba23-1d1000e6ca00","created_at":"2026-09-07T12:50:20.79745+00:00","display_order":1,"custom_summary":"OpenAI's assistant platform receiving GPT-6 Astra through a staged rollout.","prompt_examples":[],"recommended_uses":[],"relationship_type":"featured"}],"slug":"gpt-6-astra-what-changed"}] 5b:["$","div",null,{"className":"mt-8 pt-8 border-t","children":["$","p",null,{"className":"text-sm text-muted-foreground","children":["Source:"," ",["$","a",null,{"href":"https://openai.com/index/gpt-6-astra/","target":"_blank","rel":"noopener noreferrer","className":"text-teal-ink hover:underline","children":"openai.com"}]]}]}] 5c:["$","div",null,{"className":"bio","children":[["$","span",null,{"className":"mono-av","aria-hidden":"true","children":"A"}],["$","div",null,{"children":[["$","h4",null,{"children":"Assaf Biton"}],["$","div",null,{"className":"role","children":"Founder, AI Deck"}],["$","p",null,{"children":"I test AI tools by hand, write about them in Hebrew, and reject most of them. If I recommended something, it's because I used it — not because someone paid."}]]}]]}] 51:["$","nav",null,{"aria-label":"Breadcrumb","className":"flex items-center gap-2 text-sm text-muted-foreground mb-6","children":["$L67",[["$","div","/articles",{"className":"flex items-center gap-2","children":[["$","svg",null,{"ref":"$undefined","xmlns":"http://www.w3.org/2000/svg","width":24,"height":24,"viewBox":"0 0 24 24","fill":"none","stroke":"currentColor","strokeWidth":2,"strokeLinecap":"round","strokeLinejoin":"round","className":"lucide lucide-chevron-right h-4 w-4 rtl:rotate-180 text-muted-foreground/50","aria-hidden":"true","children":[["$","path","mthhwq",{"d":"m9 18 6-6-6-6"}],"$undefined"]}],"$L68"]}],["$","div","/categories/ai-news",{"className":"flex items-center gap-2","children":[["$","svg",null,{"ref":"$undefined","xmlns":"http://www.w3.org/2000/svg","width":24,"height":24,"viewBox":"0 0 24 24","fill":"none","stroke":"currentColor","strokeWidth":2,"strokeLinecap":"round","strokeLinejoin":"round","className":"lucide lucide-chevron-right h-4 w-4 rtl:rotate-180 text-muted-foreground/50","aria-hidden":"true","children":[["$","path","mthhwq",{"d":"m9 18 6-6-6-6"}],"$undefined"]}],"$L69"]}],["$","div","/articles/gpt-6-astra-what-changed",{"className":"flex items-center gap-2","children":[["$","svg",null,{"ref":"$undefined","xmlns":"http://www.w3.org/2000/svg","width":24,"height":24,"viewBox":"0 0 24 24","fill":"none","stroke":"currentColor","strokeWidth":2,"strokeLinecap":"round","strokeLinejoin":"round","className":"lucide lucide-chevron-right h-4 w-4 rtl:rotate-180 text-muted-foreground/50","aria-hidden":"true","children":[["$","path","mthhwq",{"d":"m9 18 6-6-6-6"}],"$undefined"]}],["$","span",null,{"className":"text-foreground font-medium","aria-current":"page","children":"GPT-6 Astra Is Here: What Changed and How to Use It"}]]}]]]}]