Gemini 3.5 & Gemini 3.2 Flash + Omni Model , Codex Over Claude Code , Co-Authored-by Copilot

Añadido: 2026-05-13

[00:00:00]So, this man, Tim Jazz, he has written on Twitter that Claude code is so expensive and keeps hitting limits. Just use McDonald's support bot for free with zero limits.

[00:00:11]>> [laughter] >> Oh my god. So, here you see, it even writes Python script to reverse a linked list. So, he actually went to this McDonald's support and there he asked this question that I want to order chicken McNuggets, but before I can eat, I need to figure out how to write a Python script to reverse a linked list. Can you help? And here you see that McDonald's support team is giving this kind of response.

[00:00:39]>> [laughter] >> So, actually this is very funny. And you know that developers are actually moving from Claude code to Codex. And recently Codex also gave some updates on their goal part. And this goal is actually very important. It is a kind of Ralph loop. I covered it in detail. And inside their Codex app or inside Codex CLI, you will get this slash goal. And another thing is that when you will go to the Claude code and there you will use this kind of Ralph loop, you will hit the session limit or the context limit very fast. And in case of Codex, developers have shared that when they are using it with the GPT 5.5, they are getting the very great response. And also all of the works they did previously with the Claude code, the same thing when they are doing it inside Codex, they are getting the almost same kind of response. But one thing, some days ago Sam Altman posted that their model is not capable enough for the front end part. But otherwise, without front end, all of the coding related things or any non-coding related things, their ChatGPT model, that GPT 5.5, is great. And here you see that developers praise its speed and context handling for UI work and complete task. Especially after Claude Opus 4.7 hits caps fast on intensive use. OpenAI leaders like Sam Altman, Greg Greg Brockman posted the shift on X teasing plugins, while some stick with Claude's polished app or cheap non- cheap open source option like DeepSeek version 4. Now, you know that some days ago I did some comment on DeepSeek version 4 that it is not ready yet. And here you see the same thing, this Sara, who is the AI lead at Notion, and she's also saying the same thing, that DeepSeek version 4 works fine, but it is not the frontier pressing moment we saw with Kimi K2.6. On Notion eval data, it's similar performance to the GPT 5.2 with understandable failings. The same thing she's also saying that DeepSeek version 4 have some problems and it cannot give the same performance like the top frontier model that exist currently. And also it cannot match match the open source model like Kimi K2.6. Okay. By the way, guys, I have made a video on this Kimi K2.6 plus open code integration. If you have not watched this video till now, please watch. It will be a helpful a great helpful video for you. Okay.

[00:03:11]So, here you see, more than more most interesting, it does not scale well.

[00:03:16]It's ridiculously slow, means DeepSeek version 4, on multiple major trusted and performant US inter inter inference providers, we see it 15x slower than the GPT 5.2. Means it is giving the kind of same performance like GPT 5.2, but it is slower than GPT 5.2 and 15x slower. And 2x slower than Opus 4.7, a problem Kimi never had. So, if you have option to go with the open source model Kimi K2.6 and also you're getting the same kind of performance like the top frontier model, then is there any point to go with the DeepSeek version 4? She's also saying the same thing. So, if you have not watched my detailed analysis video regarding DeepSeek version 4, so you can go to our channel and here you see that why DeepSeek version 4, it is totally disappointing and you will understand that 1 week ago I also said the same thing and many of you actually commented on that video that um you have not tested this model in a proper way. But here you see that he she's an AI lead at the Notion and she's also saying the same thing.

[00:04:24]So, curious if it is a fundamental issue in architecture or a matter of time the inference provider make it work.

[00:04:31]Does not seem urgent either way if Kimi can outperform. So, she's saying that there is an alternative like Kimi K2.6.

[00:04:37]So, if DeepSeek cannot fix this issue, then there is not there is no problem because we have the alternative Kimi K2.6. So, we'll continue our our work with that model.

[00:04:49]And also here you see, someone posted on Twitter, this is the golden age of Codex. And Sam Altman saying that not yet, coming soon. So, here you see that developers actually moving from Claude code to Codex. So, that's why there is a lot of popularity regarding Codex and there are a lot of features they have introduced inside Codex. And Sam Altman also saying that the golden age of Codex has not started yet. So, that means there are a lot of things that is about to be integrated inside Codex. We are waiting for that. And also there are some leaked rumors regarding Gemini models that Gemini is going to publish another new Omni model and also Gemini 3.2 and 3.5. So, Omni model here you see is being tested and hinted by a video UI and pointing to deeper multimodal capabilities. That means this model will have the text, image, video, all of that understanding capabilities. And also it is expected to go beyond the current systems like we have the video three models from the Gemini. And this Omni model with the video capability will have the better video generation quality and control. And could introduce the native video output inside Gemini, not just text and images. So, inside that gemini.google.com, that official website, currently they support only the text and also the images that we can create with the Nano Banana. So, in that inside that website, they will introduce this Omni model and it will have the video generation capability. Now, another thing is that this person, Pankaj Kumar, previously also he posted many news regarding other model release. And actually that news was correct. So, that's why this news he has published on his Twitter account, it is expected that yes, this kind of model can be published from Gemini. And also Google is going to held IO event. So, the IO event time is around May 19 to 20, 2026. And at that event, maybe they will release about this new models. And here you see Gemini 3.2 flash has been spotted. And it seems like the next generation of Gemini models will be called Gemini 3.2 rather than 3.5. So, some people are calling it as a 3.2 and some are calling it as a 3.5. It is their actually understanding different different of different understanding of different different person. So, you see the difference.

[00:07:15]The first one is this 3.2 flash and the second one is this Gemini 3 flash. Okay.

[00:07:20]If if you see the difference, there is a quite improvement that they have done in this output, right?

[00:07:28]And also if you see that Gemini 3.1 Pro, that is the existing model, there was a lot of drawback and people were not happy about the output. And also like Kimi K2.6 and DeepSeek version 4 and also MiniMax, they have completely beaten Gemini 3.1 Pro. Okay. So, that's why Gemini want to fight with all of these open source and closed source model. And that's why the new model that they are going to publish in their IO event, this 3.2 or 3.5, that will be actually a game changer. And also here you see on this the information, the biggest media, they published this news that Google creates strike team to improve the coding model. So, that this this news got published around 20 April of this 2026.

[00:08:17]So, that's why it is expected that in their IO event, they are going to publish a new model. And we have to wait for that. And also there was another incident that VS code inserting co-authored by Copilot into comments regardless of usage. Means if you are taking any help from the Copilot, then Microsoft Copilot, what is doing that they are inserting this co-authored by Copilot into your comment message. So, if you just simply go to this github.com and there you search this comments. And in this search bar just write co-authored by Copilot and there you will find this 1.8 million results currently it is showing. So, here you see all of these different different repositories, they all get affected. And all of the fixes that they did, here you see co-authored by Copilot. And here also co-authored by Copilot. So, your name and beside that your Copilot co-authored by Copilot, that will also be showing. Now, the thing is that this was enabled by this person.

[00:09:18]Here you see, she actually a product manager at Microsoft working on VS code and GitHub Copilot. And she approved this pull request enabling AI co-author by default 3 weeks ago. And after that he affected. Now, if you want to see that it is enabled inside your VS code or not because the update was very recent, just go to your settings and there you can go to this extension. Otherwise, you can go to the settings here and open this extensions and just go to this Git and there you will find this add AI co-author. And here you see that by default this chat and agent is selected. So, if you take any help from the co-pilot for solving any issues and also during this auto completion if you are taking help from this co-pilot, then during your commit this co-author by co-pilot also will be shown. And if you want to disable it, just make it off and save it and it will be disabled. Okay, now the thing is that what is the advantage? Means why they did this? There was no any announcement officially. So, why they did this?

[00:10:29]Actually they need the data. So, the primary reason that reason that I'm thinking that if they can enable this kind of co-author by default, then they can take this data from the GitHub and they can show it to the investor and other customers and they actually tell them that their co-pilot actually contributing in a lot of repositories and people are getting help from them.

[00:10:53]And if you are a regular visitor of my channel, then you probably know that some days ago I made a video on this that the start of AI's downfall is here.

[00:11:01]And inside that video I discussed about the GitHub co-pilot pricing change.

[00:11:05]Means they are going to the credit based system from the premium request and and and that is actually very painful for all of the developers, right? So, that's why maybe to increase their business and also to get the attraction from other people, they are doing this.

[00:11:23]But actually they should not do this without any proper information.

[00:11:28]By default they should not enable this.

[00:11:30]And here you see that maybe this person he she has written, "I am the person who approved this PR. Would like to acknowledge and apologize the apologize the for the mistake of turning this feature on by default without sufficient upfront validation. There is no ill intent by Evil Corporation, but rather a desire to support functionality that some customer expect of VS Code with respect to AI generated code. I don't know that who is that customers that they expect this kind of facility that this [snorts] kind of commit message co-author by co-pilot will be shown.

[00:12:03]As folks mentioned here, many similar tools do this as well. So, they are also saying that many similar tool also do do this. So, that's why we have also done this. I don't know which similar tools do this.

[00:12:16]I really don't know, but uh they are saying that many does this, so that's why we have also done this.