[{"data":1,"prerenderedAt":14},["ShallowReactive",2],{"artikel-en-berita-ai-gemini-4-argon-gpt-6-beam":3},{"slug":4,"lang":5,"title":6,"description":7,"date":8,"updated":9,"topic":10,"product":9,"cover":11,"readMin":12,"html":13},"berita-ai-gemini-4-argon-gpt-6-beam","en","AI news this week: Gemini 4 Argon, faster GPT-6 and the open Beam model","Google announced Gemini 4 Argon but limited access, OpenAI sped up GPT-6 Astra and GPT-6.1 Sol for subscribers, and Reflection introduced Beam. Here is the summary and what it means for you.","2026-10-06",null,"berita","\u002Fimages\u002Fartikel\u002Fberita-ai-oktober-2026.webp",4,"\u003Cp>The first week of October 2026 was a busy one in AI. Three big stories landed within a few days: a new Google model the public cannot use yet, ChatGPT answering faster, and a new open model from the United States.\u003C\u002Fp>\n\u003Cp>Here is a plain language summary of each, and what it means for everyday users like us.\u003C\u002Fp>\n\u003Ch2>1. Gemini 4 Argon: Google's most capable model, but not open yet\u003C\u002Fh2>\n\u003Cp>On 30 September 2026, Google announced \u003Cstrong>Gemini 4 Argon\u003C\u002Fstrong>, its most powerful AI model so far. According to Google, Argon is built for deep reasoning across long, complex work such as software engineering, legal and finance work, and cybersecurity defense.\u003C\u002Fp>\n\u003Cp>The interesting part: Google is \u003Cstrong>not releasing it to the public\u003C\u002Fstrong>. For now, Argon is only available to a vetted group of cybersecurity defenders through the \u003Cstrong>Fairwind Program\u003C\u002Fstrong>, plus Google's own internal teams. Approved organisations must follow strict rules, including phishing resistant two factor authentication, and only their security teams may use it.\u003C\u002Fp>\n\u003Cp>Why so careful? Argon is very good at finding and fixing software vulnerabilities. The same skill could be misused by attackers to target banks, hospitals or government systems, so Google chose a phased release.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>When can you use it?\u003C\u002Fstrong> Google says paid API customers and \u003Cstrong>Google AI Ultra\u003C\u002Fstrong> subscribers are next in line. No date has been announced.\u003C\u002Fp>\n\u003Ch2>2. GPT-6 Astra and GPT-6.1 Sol are now faster\u003C\u002Fh2>\n\u003Cp>On 5 October 2026, OpenAI's Thibault Sottiaux announced that the default speed of \u003Cstrong>GPT-6 Astra\u003C\u002Fstrong> and \u003Cstrong>GPT-6.1 Sol\u003C\u002Fstrong> had been increased by about 50% for ChatGPT subscribers.\u003C\u002Fp>\n\u003Cp>In simple numbers, the models now write at roughly \u003Cstrong>50 tokens per second\u003C\u002Fstrong>, up from about 30. You do not need to change any settings. The change applies to OpenAI products such as ChatGPT and \u003Cstrong>Codex\u003C\u002Fstrong>, as well as partner apps that use Sign in with ChatGPT.\u003C\u002Fp>\n\u003Cp>A little context:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>GPT-6 Astra\u003C\u002Fstrong> is OpenAI's flagship model for demanding reasoning and coding.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>GPT-6.1 Sol\u003C\u002Fstrong> is the lower cost option for coding and professional work.\u003C\u002Fli>\n\u003Cli>Faster here means the answer text \u003Cstrong>appears on screen sooner\u003C\u002Fstrong>. It does not necessarily mean smarter answers, or that long tasks finish twice as fast.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>This was also day one of a 28 day pledge from OpenAI to ship one improvement every day for Codex and work users.\u003C\u002Fp>\n\u003Ch2>3. Reflection Beam: a new open model from the US\u003C\u002Fh2>\n\u003Cp>Also on 5 October 2026, the startup \u003Cstrong>Reflection AI\u003C\u002Fstrong> introduced \u003Cstrong>Beam\u003C\u002Fstrong>, its first open model. Reflection was founded by two former Google DeepMind researchers.\u003C\u002Fp>\n\u003Cp>An open weight model means anyone can download it and run it on their own servers, without depending on the company that made it. That matters for businesses and government agencies that want to keep control of their data.\u003C\u002Fp>\n\u003Cp>A few facts about Beam:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Focused on \u003Cstrong>coding, reasoning and AI agent tasks\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>Can read up to \u003Cstrong>1 million tokens\u003C\u002Fstrong> of context at once\u003C\u002Fli>\n\u003Cli>Reflection claims it performs on par with leading Chinese open models such as GLM-5.2, while using \u003Cstrong>3 to 4 times less compute\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>The model weights will be released \u003Cstrong>later in October\u003C\u002Fstrong> under the Apache 2.0 licence, which allows commercial use\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Keep in mind that these performance figures come from Reflection itself and have not been independently verified. We will know how it really performs once the weights are out and other developers test it.\u003C\u002Fp>\n\u003Ch2>What it means for you\u003C\u002Fh2>\n\u003Cp>If you use AI for writing, content creation or everyday work:\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>Paid ChatGPT users\u003C\u002Fstrong> will simply notice answers arriving faster, with nothing to do.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Gemini 4 Argon cannot be tried yet.\u003C\u002Fstrong> If anyone offers you access to Argon right now, be careful. It may be a scam.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Open models like Beam\u003C\u002Fstrong> are more relevant to businesses and developers for now. But this competition is good for everyone, because it usually pushes prices down and quality up for the AI tools we use every day.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>Want to know which AI tools suit your work right now? See my picks on the \u003Ca href=\"\u002Fen\u002Fai-tools\">AI Tools\u003C\u002Fa> page.\u003C\u002Fp>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fblog.google\u002Finnovation-and-ai\u002Fmodels-and-research\u002Fgemini-models\u002Fgemini-4-argon\u002F\" target=\"_blank\" rel=\"noopener\">Introducing Gemini 4 Argon\u003C\u002Fa> (Google)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fdeepmind.google\u002Ffairwind-program\u002F\" target=\"_blank\" rel=\"noopener\">Fairwind Program\u003C\u002Fa> (Google DeepMind)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fruntimewire.com\u002Farticle\u002Fopenai-gpt-6-models-speed-optimization-tibo-sottiaux\" target=\"_blank\" rel=\"noopener\">OpenAI says GPT-6 Astra and Sol now stream 50% faster\u003C\u002Fa> (Runtimewire)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Ftechcrunch.com\u002F2026\u002F10\u002F05\u002Freflection-debuts-beam-a-open-weight-ai-model-to-rival-chinese-models-at-lower-compute-cost\u002F\" target=\"_blank\" rel=\"noopener\">Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost\u003C\u002Fa> (TechCrunch)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Freflection.ai\u002Fblog\u002Fintroducing-beam\" target=\"_blank\" rel=\"noopener\">Introducing Beam\u003C\u002Fa> (Reflection)\u003C\u002Fli>\n\u003C\u002Ful>\n",1791263603903]