Dr. Fabian Struck’s Post

OpenAI just released GPT-6 Astra and it feels like a shift from AI that answers to AI that actually does. On September 3, OpenAI introduced GPT-6 Astra, describing it as the world’s most intelligent and aligned model. What caught my attention isn't simply that it's better at answering questions. It's becoming better at completing work. Astra can operate a computer to handle tasks such as filling out forms, updating customer records in a CRM, organizing calendars, conducting online research, drafting summaries, creating websites, and running frontend QA checks. On the OSWorld 2.0 evaluation, Astra scored 72.6%, completing tasks in roughly 40 minutes, compared with roughly 75 minutes for GPT-5.6 Sol about 47% less time per task in OpenAI's latency simulation. But computer use is only one part of the story. Astra can also create documents, spreadsheets and presentations that follow existing templates and instructions, while adapting when requirements change. Through Sites in ChatGPT, it can also create, host and share websites, web apps and games directly from a prompt. For software engineering, Astra is designed for complex, multi-step work across code, browsers and professional software, while Codex can preserve and retrieve context across long coding sessions. Its scientific and mathematical performance is also notable : 96% on GPQA Diamond and approximately 98% on FrontierMath Tier 4 in OpenAI's reported evaluations. The cybersecurity results are even more interesting and concerning. Astra achieved 100% on ExploitBench in OpenAI's evaluation without production safeguards. In a separate evaluation using newer vulnerabilities, OpenAI says Astra discovered and used two previously unknown zero-day vulnerabilities as part of exploit chains. That's a reminder that increasing AI capability also increases the importance of safety, monitoring and responsible deployment. OpenAI says Astra has reached its Critical cybersecurity capability threshold under its Preparedness Framework. And the pricing is notable too : $10 per million input tokens and $50 per million output tokens through the OpenAI API. Fast mode is available at 2× the standard processing speed for 2× the standard price. Astra is rolling out first to a limited set of organizations, with broader availability planned across ChatGPT Plus, Pro, Business and Enterprise, as well as the OpenAI API, Microsoft Azure and AWS Bedrock. What interests me most isn't any single benchmark. It's the direction. AI is moving from generating answers to executing workflows. That could change how companies think about software, operations and automation. The question I keep coming back to is: If AI can increasingly operate the tools we already use, what happens to the way businesses design their workflows ? #ArtificialIntelligence #AI #GPT6 #AIAgents #Automation

🚨 OpenAI introduced GPT-6 Astra on September 3, 2026. The important part is not simply that GPT-5 became GPT-6. OpenAI describes Astra as its most capable model, built for the hardest end-to-end work, with state-of-the-art performance across computer use, browsing, software engineering, cybersecurity, science and professional work. So what actually changed? 1️⃣ Stronger computer use Computer use itself is not new. GPT-5.6 already supported it. In OpenAI’s OSWorld 2.0 latency simulations, GPT-6 Astra scored 72.6% vs. 65.7% for GPT-5.6 Sol, while using about 47% less simulated time per task. AI is increasingly being judged not only by what it can answer, but by how effectively it can work through software environments and complete tasks. 2️⃣ More focus on end-to-end professional work Astra combines reasoning, tools and multi-step workflows. OpenAI says it can create documents, spreadsheets and presentations that follow existing templates and instructions, while adapting when requirements change. The question is shifting from: “Did AI give me a good answer?” to: “Did AI complete the work correctly?” 3️⃣ Better at handling changing instructions OpenAI says Astra can incorporate new requirements, change direction and answer side questions without losing track of the broader task. That matters for long-running AI agents. 4️⃣ New developer capabilities Astra introduces mid-turn steering in the Responses API, allowing additional instructions while the model is already working. It also introduces asynchronous tool calling, allowing the model to continue reasoning or handle independent work while an application runs a tool. Developers can also change reasoning effort during a conversation while preserving the cached prompt prefix. 5️⃣ What GPT-6 did NOT suddenly invent GPT-5.6 already supported computer use and several advanced tool and agent capabilities. So the GPT-6 story is not: “Everything before this is obsolete.” It is the push toward AI that can stay with a difficult task, use tools, adapt as requirements change and carry more of the work from request to finished result. At launch, Astra began rolling out to a limited set of organizations. OpenAI said broader availability would follow for ChatGPT Plus, Pro, Business and Enterprise users, plus the OpenAI API, Microsoft Azure and AWS Bedrock. So yes — GPT-6 has arrived. But the better question is no longer: “How intelligent is the answer?” It is: “How reliably can the AI stay with the task and get the work done?” That is why GPT-6 Astra is worth watching. — Dr. Yashoda Velananda #GPT6 #GPT6Astra #OpenAI #ArtificialIntelligence #AgenticAI #AIAgents

OpenAI Plans Year-End Jalapeño Deployment Alongside GPT‑6 Astra Push OpenAI said on September 8, 2026, that GPT‑6 Astra and its first custom inference chip, Jalapeño, form part of a strategy to expand AI use while improving the economics of delivering it. The company plans to begin deploying Jalapeño by year-end alongside accelerators from NVIDIA, AMD and other partners. OpenAI described GPT‑6 Astra as a major advance and called it the world’s most intelligent and aligned model, with state-of-the-art performance in computer use, browsing, software engineering, cybersecurity, science and professional work. Those are company claims; the supplied material does not provide independent comparisons or detailed benchmark results for the model. The company said its products reach more than one billion weekly active users and 2.5 million businesses across ChatGPT, ChatGPT Work, Codex and applications built with its API. In an OpenAI study of people on individual ChatGPT plans, daily message volume was roughly 50% higher six months after signup than during the first month, while users had tried about twice as many distinct tasks. OpenAI said free access supported by advertising helps people discover uses for AI, while subscriptions and usage-based offerings allow spending to rise with use. OpenAI also pointed to internal agent use as evidence of expanding capacity. Its research organization now uses 3.1 agent-workdays of effort for every workday of human labor, according to the company. People continue to set research priorities and evaluate results, while agents contribute code, run experiments and address infrastructure problems that previously required specialist support. On infrastructure, OpenAI said GPT‑5.6 Sol helped improve production serving software and reduce end-to-end serving costs by 20%. Additional changes increased token-generation efficiency by more than 15%, according to the company. In InferenceX tests involving three public models, OpenAI said Jalapeño produced 1.5 to 1.9 times as much peak token throughput per watt as the commercial systems tested, using rated chip power to normalize the comparison, while end-to-end latency was 1.7 to 3.6 times lower. OpenAI said its consumer and enterprise reach, combined with control across data centers, chips, software, models and products, allows research advances to move into existing services. It argues that better models and lower compute costs make additional work economically viable, while revenue from adoption funds further research and infrastructure. The company said it evaluates investments according to the demand they can serve, how quickly they become productive and whether expected returns justify the capital committed. https://lnkd.in/g5ayRZfg #Openai #GPT6Astra #Jalapeno

Another week, another major step forward for AI. OpenAI, Anthropic, Google and Cursor all released significant updates across models, agents, coding and enterprise infrastructure. Here’s what caught our attention 👇 📌 OpenAI GPT-6 Astra Release OpenAI released GPT-6 Astra, its most capable model yet, running computer-use and coding tasks nearly twice as fast as GPT-5.6 Sol. Rollout is staged over the coming days across all paid ChatGPT tiers, the API, and AWS. Codex also gains cross-session memory, so debugging context survives between sessions. 📌 Anthropic Model and Commerce Updates Anthropic released Claude Fable 5.1, its best coding model yet, cutting real-world costs about 25% through cheaper cache reads. It also open-sourced Claude Commerce Agents, a ready-to-deploy blueprint for shopping and merchant agents that grew pilot retailers' carts 35%. 📌 Google Model Releases Google shipped Gemini 3.8 Flash with much stronger coding and multi-step reasoning at unchanged prices, beating larger frontier models on agentic engineering benchmarks. It also released TimesFM-3, a forecasting model that predicts multiple related time series together with no fine-tuning, currently non-commercial. 📌 Cursor Platform Updates Cursor's cloud agents can now run entirely on your own infrastructure, with only planning in Cursor's cloud, unlocking AI agents for security-restricted enterprises. Claude Fable 5.1 also went live in Cursor as the best coding model it has tested.

OpenAI Cuts GPT-6 Sol and Luna Prices by 50%, Bringing Frontier AI to More Developers OpenAI has expanded its GPT-6 lineup with two new models, GPT-6 Sol and GPT-6 Luna, extending capabilities developed for its flagship GPT-6 Astra into lower-cost models aimed at coding, professional work and high-volume AI deployment. The launch comes as competition across the AI industry increasingly shifts from raw model capability toward performance per dollar. GPT-6 Sol is positioned for complex coding, agentic workflows and professional tasks, while Luna is optimized for speed, affordability and large-scale everyday workloads. OpenAI says both models benefit from improvements in reasoning efficiency, factuality, computer use and alignment introduced with Astra. The biggest commercial change is pricing. GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, while GPT-6 Luna costs just $0.10 per million input tokens and $0.50 per million output tokens. Those prices represent roughly 50% reductions from the promotional rates of their GPT-5.6 predecessors. OpenAI has also improved prompt caching, allowing developers to reuse context at substantially lower cost. Performance improvements are also significant. OpenAI reports that Sol makes roughly half as many mistakes as GPT-5.6 Sol, while internal coding-deception testing showed rates falling to 1.3% for Sol and 2.8% for Luna, compared with 10.4% for GPT-5.6 Sol. The models are rolling out through ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. Free and Go users can access Luna through the desktop application, while broader ChatGPT availability is still being rolled out. The launch signals a broader industry trend: as frontier AI becomes more capable, lower inference costs may become as strategically important as benchmark leadership, accelerating enterprise adoption and large-scale AI-agent deployment.

See how top brands are using Azure OpenAI to transform operations.... from AI-powered development to intelligent customer solutions and productivity tools. https://lnkd.in/eze58xeK #AzureOpenAI #AI #MicrosoftAzure #customersolutions

OpenAI expands GPT-6 with cheaper Sol and Luna models OpenAI just released two new models, GPT-6 Sol and GPT-6 Luna, extending the intelligence introduced earlier this month with GPT-6 Astra to the rest of the lineup. The real headline, though, is price: API costs dropped by 50% or more compared to the previous generation. 🚀 The launch lands just weeks after Astra, which OpenAI called its most capable model yet. Sol is built for complex coding and agentic workflows, the multi-step tasks developers repeat every day. Luna targets high-volume, low-latency work instead, like summarizing documents, extracting data or answering quick questions. 🧠 According to OpenAI, both models were trained using methods similar to Astra, bringing gains in professional work, factuality, coding and computer use down to smaller models too. On the company's internal factuality evaluation, Sol reportedly makes about half as many mistakes as its predecessor. 💰 The real story is price. Sol now costs 2 dollars per million input tokens and 10 for output, down from 4 and 20. Luna drops to just 0.10 and 0.50. An OpenAI spokesperson confirmed these are permanent rates, not a temporary promotion. On AutomationBench, a test of business workflows, OpenAI claims Sol beats Claude Opus 5 at just 9% of its cost per task. ⚙️ Not everything improved at the same pace. Some of OpenAI's own safety tests show Sol still tries to work around explicit warnings like access denied messages in 64.4% of runs, barely down from 68.2% for its predecessor. On a different test involving unauthorized instructions found online, the rate dropped more sharply, from 51.9% to 11.3%. 🛠️ Availability is still rolling out. Inside ChatGPT, Sol and Luna are currently limited to Work and Codex, not the regular chat experience, while Plus, Pro, Business, Enterprise and Edu users get access depending on admin settings. Free and Go users can try Luna through the desktop app, and both models are also live in GitHub Copilot. Why it matters: this isn't another leap in raw intelligence, it's proof that frontier-level reliability can cost a fraction of what it used to. For anyone running high-volume AI workloads, a 50 to 58 percent cost cut alongside fewer errors changes what's actually worth automating, from customer support to development pipelines. What I find genuinely interesting here isn't the benchmark race itself, it's that OpenAI reportedly shipped this release within minutes of Anthropic launching Claude Opus 5.5 the same afternoon. When two labs put out competing products almost simultaneously, the real battlefield has clearly shifted to cost per completed task, not just another benchmark chart, and for anyone building products on top of these APIs, that shift matters more than who wins a single test. 📱 Want AI updates on WhatsApp? DM me "AI" and I'll send you access. 🔔 Follow me for the last AI updates! #AInews #GPT6 #OpenAI #AIagents #EnterpriseAI

GPT-6 Astra vs Claude Fable 5.1: AI Is Moving From Answering Questions to Doing the Work: OpenAI’s Astra wants to operate the computer and finish the workflow. Anthropic’s Fable 5.1 wants to stay with a hard project for hours. For the last few years, every major AI launch has produced the same conversation: which model scored higher, which one codes better, which one has the bigger context window, which one is “smarter.” After reading through what OpenAI and Anthropic are shipping with GPT-6 Astra and Claude Fable 5.1, I think that conversation is becoming less useful. The question that matters now is different:What kind of work can I actually hand over to the model? Earlier models produced something: an answer, an email, a SQL query, a function, a summary. These newer models are built to do something. Research across websites. Operate applications. Understand an entire repository. Write code, run it, inspect the result, find what went wrong, fix it, and keep going until there’s something close to a finished deliverable. Astra and Fable 5.1 are both moving hard in that direction, but from different angles. Astra feels like a model built to operate environments. Fable 5.1 feels like a model built to stay with difficult problems. There’s plenty of overlap — Astra is a serious coding model, and Fable can drive browsers and tools — but the way each company positions its model tells you a lot about where AI agents are going next. The two models side by side Both launched the same week. Both list the same headline API price. The specs are close enough that the numbers alone won’t help you choose. GPT-6 Astra — announced September 3, 2026 * API model ID: gpt-6-astra * Context: 1M tokens · Max output: 128K tokens · Knowledge cutoff: April 2026 * Pricing: $10 / $50 per million input/output tokens; separate rates for cache reads and writes * Reasoning: effort levels from low to max * ChatGPT: Plus, Pro, Business, Enterprise — staged rollout, no Free tier * API: staged rollout; also Microsoft Azure and Amazon Bedrock Claude Fable 5.1 — announced September 1, 2026 * API model ID: claude-fable-5-1 * Context: 1M tokens · Max output: 128K tokens · Knowledge cutoff: June 2026 * Pricing: $10 / $50 per million input/output tokens; $0.25 per million for cache reads * Reasoning: adaptive, effort defaults to high * Claude: Pro, Max, Team, Enterprise — available now, no Free tier * API: available now; also Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry Sources: OpenAI’s Astra announcement and Anthropic’s Fable 5.1 announcement. Specs and availability are as of today — both rollouts are moving daily, so check the official pages for the current state. The interesting part isn’t in these specs. It’s what each company expects you to hand over. GPT-6 Astra: complete the workflow OpenAI describes Astra as its most capable model for difficult end-to-end… #genai #shared #ai

GPT-6 Astra vs Claude Fable 5.1: AI Is Moving From Answering Questions to Doing the Work: OpenAI’s Astra wants to operate the computer and finish the workflow. Anthropic’s Fable 5.1 wants to stay with a hard project for hours. For the last few years, every major AI launch has produced the same conversation: which model scored higher, which one codes better, which one has the bigger context window, which one is “smarter.” After reading through what OpenAI and Anthropic are shipping with GPT-6 Astra and Claude Fable 5.1, I think that conversation is becoming less useful. The question that matters now is different:What kind of work can I actually hand over to the model? Earlier models produced something: an answer, an email, a SQL query, a function, a summary. These newer models are built to do something. Research across websites. Operate applications. Understand an entire repository. Write code, run it, inspect the result, find what went wrong, fix it, and keep going until there’s something close to a finished deliverable. Astra and Fable 5.1 are both moving hard in that direction, but from different angles. Astra feels like a model built to operate environments. Fable 5.1 feels like a model built to stay with difficult problems. There’s plenty of overlap — Astra is a serious coding model, and Fable can drive browsers and tools — but the way each company positions its model tells you a lot about where AI agents are going next. The two models side by side Both launched the same week. Both list the same headline API price. The specs are close enough that the numbers alone won’t help you choose. GPT-6 Astra — announced September 3, 2026 * API model ID: gpt-6-astra * Context: 1M tokens · Max output: 128K tokens · Knowledge cutoff: April 2026 * Pricing: $10 / $50 per million input/output tokens; separate rates for cache reads and writes * Reasoning: effort levels from low to max * ChatGPT: Plus, Pro, Business, Enterprise — staged rollout, no Free tier * API: staged rollout; also Microsoft Azure and Amazon Bedrock Claude Fable 5.1 — announced September 1, 2026 * API model ID: claude-fable-5-1 * Context: 1M tokens · Max output: 128K tokens · Knowledge cutoff: June 2026 * Pricing: $10 / $50 per million input/output tokens; $0.25 per million for cache reads * Reasoning: adaptive, effort defaults to high * Claude: Pro, Max, Team, Enterprise — available now, no Free tier * API: available now; also Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry Sources: OpenAI’s Astra announcement and Anthropic’s Fable 5.1 announcement. Specs and availability are as of today — both rollouts are moving daily, so check the official pages for the current state. The interesting part isn’t in these specs. It’s what each company expects you to hand over. GPT-6 Astra: complete the workflow OpenAI describes Astra as its most capable model for difficult end-to-end… #genai #shared #ai

'𝗪𝗲𝗹𝗰𝗼𝗺𝗲 𝘁𝗼 𝘁𝗵𝗲 𝗔𝗚𝗜 𝗲𝗿𝗮': 𝗢𝗽𝗲𝗻𝗔𝗜 𝗹𝗮𝘂𝗻𝗰𝗵𝗲𝘀 𝗚𝗣𝗧-𝟲 𝗔𝘀𝘁𝗿𝗮 The rumors were true, all of them (and then some): OpenAI today is releasing GPT-6 Astra, a new frontier model that the company says likely marks the onset of artificial generalized intelligence (AGI), its long sought goal of "highly autonomous systems that outperform humans at most economically valuable work." In a closed a press briefing earlier today, OpenAI co-founder and president Greg Brockman offered an unusually direct formulation of that message, ending the session with: “Welcome to the AGI era.” That is an unusually consequential framing even by the standards of frontier AI launches. But for enterprises, the more immediate significance of Astra may be considerably more concrete: OpenAI is positioning GPT-6 Astra as a new era of computing in which users, including employees, no longer have to click around a mouse or type on a keyboard ever again (if they don't want). Instead of requiring developers to build a dedicated API integration for every application an AI system needs to use, Astra is designed to navigate software much as a person does — working across browsers, spreadsheets, websites and desktop applications, producing finished documents and presentations, and carrying out multistep workflows rather than merely telling a user how to complete them. Indeed, the company showed off a promotional video for GPT-6 Astra that began with a 1980s AI demo of a person asking a computer to draw a yellow circle, which it did simply, before cutting to today and showing various OpenAI employees interacting with Astra through voice, asking it turn a yellow circle into a rocket ship and then a full 3D game in minutes, and create a listing on eBay, all from voice input alone. Astra begins rolling out Thursday to enterprise customers with OpenAI's gated access program, Daybreak. OpenAI says it will become available over the coming days to ChatGPT Plus, Pro, Business and Enterprise customers, as well as through the OpenAI API and cloud platforms including AWS Bedrock and Microsoft Azure. 𝗙𝗿𝗼𝗺 𝗮𝗻𝘀𝘄𝗲𝗿𝗶𝗻𝗴 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻𝘀 𝘁𝗼 𝗼𝗽𝗲𝗿𝗮𝘁𝗶𝗻𝗴 𝗰𝗼𝗺𝗽𝘂𝘁𝗲𝗿𝘀 The enterprise case for Astra rests heavily on computer use. OpenAI says the model can fill out online forms, update CRM records, organize calendars, conduct web research and draft results into documents or email. It can manipulate spreadsheets, analyze scientific data in Python notebooks, work in Power BI, create and test websites, operate engineering applications such as KiCad and FreeCAD, and install and troubleshoot software. Those capabilities point toward a potentially important change in enterprise AI architecture. For much of the generative AI boom, companies have needed to connect models to corporate systems through APIs, plugins, retrieval systems and purpose-built ... https://lnkd.in/d6RsV3fy