Claude Might Be Going In A Different Direction + More

This past week, the AI updates have been absolutely insane with OpenAI dropping a new model that fixes everything people hated about GPT5 and Google leaking updates so powerful that we can't wait to test them out ourselves. Anthropic also reported some shocking AIEL activities that made us question the potential of AI along with their interesting experiments testing Claude's coding abilities and potential in other fields. Hey everyone, if you're new here, this is AI Labs and welcome to another episode of Context Weekly, where each episode informs you about everything that has happened in the past week with AI. As mentioned in our previous weekly video, OpenAI was planning to release a new model. This week, this new model GPT 5.1 began rolling out to users on November 12th. You already know that GPT5 received criticism for providing worse answers compared to previous models. So, OpenAI released two upgrades. GPT 5.1 instant, a more conversational and warmer model, and the thinking model, which offers advanced reasoning and performs better on complex tasks. By default, chat GPT selects GPT 5.1 instant, but you can switch to thinking or other models as needed. Many users have reported that this model is significantly better in conversational style, more empathetic in tone, and far better at following instructions than the previous version. We might release a deep dive video on this model, so look out for that. OpenAI also released another update with the launch of GPT5 CEX Mini in Codex. This model is said to be the smaller and most cost-effective version of GPT5 Codex, providing up to four times more usage as part of your subscription. So whenever you reach 90% of your usage limit, you will automatically be switched to this model for further use. You can actually go ahead and try it out in Codeex right now. GitHub also announced that GPT 5.1 Codeex and its mini version are in public preview on the Copilot. Next up, Google has launched some huge updates for Gemini. Starting off with the first and biggest of them all is Gemini 3. As we mentioned in the previous video, Gemini 3 was probably going to be coming soon and now we have more information about it. This model was leaked to multiple testers and they were amazed by its capabilities. For example, someone asked it to clone Mac OS in the browser and it delivered a near-perfect version with just one prompt, exactly replicating all the common apps. People are also talking about the model's front-end capabilities, which are considered pretty impressive compared to previous versions. Gemini models were already considered really good, but this new update is expected to be a huge upgrade for UI design. Before we move on to the next update, here's a word from our sponsor, Scalekit. Every B2B AI founder knows the pain of dealing with passwords. They seem simple, but they create real friction. Users drop off during signup. Support teams get buried under reset requests and security risks never fully disappear. That's why modern B2B AI products are shifting to passwordless authentication. If you want smoother onboarding and stronger security without adding engineering overhead, Scalekit is worth checking out. It gives you plug-and-play modules for magic links, pass keys, and even one-click SSO for enterprise customers. You can also add social login to speed up user signups. The best part, Scalekit works with your existing setup. No refactoring, no migrations. Its modular architecture lets you drop in exactly the O features you need. If you're looking to modernize your AI apps, ScaleKit gives you a purpose-built O stack that fits seamlessly into modern AI workflows. Click the link in the pinned comment and start building with Scale Kit today. Other than that, we also have some new leaks from people who tested Nano Banana 2. The most impressive leak I found was that when given pages torn into separate pieces containing parts of the text, it was able to reassemble them and recover the original message with an accuracy no model has achieved before. This is a really cool development that could transform the landscape of AI image understanding. It was also revealed that it could generate a screenshot of a Windows 11 desktop with Chrome open displaying YouTube. It also generated a nearly accurate Google DeepMind landing page with just a simple prompt and no additional references. If it weren't for some text issues in the corner, I wouldn't have been able to tell that it was AI generated. This is huge given that the previous version of the model struggled with text. People are really excited to try out this new model and it feels like the closest we've come to actually replacing Photoshop. We also have some new confirmed releases from Google. Gemini now has multiple integrated tools and with Google Maps you can use Gemini to navigate using your voice while driving Gemini can reroute you, help you make stops and manage your travel all controlled through it. This new feature will be rolled out to iOS and Android users in the coming weeks with priority given to Android Auto support. The rollout will initially focus on the US and will later expand to other countries as well. Google Opal, Google's Noode AI app builder, which was previously limited to the US only, is now available globally across 160 countries. Google Opel is an easytouse platform similar to N8N and OpenAI's agent kit, automating tasks within your Google workflow. It helps you create apps with just a simple prompt and is fully integrated with all Google products. This is something many people in other countries have been waiting for. It's still in the experimental phase, not a completed product, and is available for free during this beta phase. So, go check it out. Google has another update that's really exciting for ads and commerce. They are releasing two agents that will change the way we work. The analytics adviser and the ads adviser. The ads adviser is your AI partner, helping you manage campaigns directly within Google Ads and maximize performance. The analytics adviser specializes in unlocking powerful insights for you. It can analyze trends, generate reports, and provide answers based on a holistic view of your data. This is going to be a real gamecher for those who are managing campaigns with Google. You can access both agents directly within your accounts early next month. In the Gemini application, there is a new feature called deep research. You can now specify the sources. And the cool thing is you can use your Gmail, your drive, and your Google chats as sources for your research process or you can ground it with information from file uploads. These are features that other companies deep research tools like open ais or clouds don't have. Based on your prompt, it creates a plan and confirms it with you. You can edit this plan as needed and then it begins the research process. It generates a properly structured research document using all the sources you specified, compiling a thorough report with all the sources linked. Next up, there was a very interesting experiment conducted by Anthropic. They had engineers with zero experience in robotics and divided them into two teams of two. Both teams were tasked with programming a robot dog to perform a set of tasks. One team using Claude and one team without. The results were impressive. The team using Claude was able to complete the tasks in significantly less time compared to the team that wasn't using Claude. Both teams ultimately completed the tasks, but the Claude assisted team finished much faster. This raises an exciting question. Is AI going to transform the field of robotics? as profoundly as it's transforming software development. In the future, it's possible that Claude could program complex human-like robots. Anthropic also released a very interesting news article. As you know, security is a major concern in AI development and with AI support, attacks are becoming more powerful. What happened was that in midepptember 2025, they detected suspicious activity that turned out to be a spying campaign. The attacker used AI's agentic abilities to carry out malicious actions. Anthropic reported that the attack was linked to Chinese state sponsored groups which manipulated Claude code to attempt infiltrating 30 global targets and they were partially successful. When this was detected, an investigation began. Cyber attacks like this exploit a model's intelligence, its agentic capabilities and access to tools. With this combination, attackers can deploy multiple pieces of malicious software such as password crackers, network scanners, and other harmful programs. What Anthropic found was that the human operator used clawed code with MCP servers to collect data and perform actions intelligently scanning for vulnerabilities. In the end, this data was used to exploit the targeted companies. This represents a major emerging threat, vibe hacking. Malicious actors can now use AI systems to amplify attacks in ways that have never been seen before in the AI landscape. Next up, IBM has launched two new quantum processors at the quantum developer conference held on November 12th. The first processor is the Quantum Nighthawk which delivers 30% higher performance than previous processors. The second is the IBM quantum loon processor which is even more powerful. IBM expects to achieve true quantum advantage by the end of 2026 with fully functional quantum computing arriving by 2029. This is particularly important for AI because quantum computers can operate at significantly higher speeds enabling the training of larger and more complex models with massive data sets. This could accelerate AI development by years. Moving on, Minecraft has also brought the power of AI agents right into your game with an agent called Steve. Instead of building block byblock, which takes a long time to create structures, you can have this AI agent build them with you. Just like with other AI agents, you can use this as your creative collaborator and let Steve do the work for you with just a simple prompt. So, that concludes the updates for this week. Keep an eye out for the next update coming next week. If you'd like to support the channel and help us keep making videos like this, you can do so by using the super thanks button below.