00:00I honestly thought that this weekend would be a bit laid back, but that is definitely
00:04incorrect because we have some huge AI headlines surfacing today.
00:09Anthropic's hyped up flagship model Mythos is appearing to be getting prepared for deployment
00:14across cloud apps like Cloud Code as well as the SDK for further testing, and some users
00:20are starting to notice traces of it appearing inside the app recently.
00:24On the topic of Anthropic, the Sonnet series is finally getting a new upgrade, fairly
00:31soon alongside another Opus refresh, as the Sonnet 4.8 and Opus 4.8 model slugs were recently
00:39discovered on Google Vertex AI.
00:41Moving over to OpenAI side of things, it looks like OpenAI is finally getting close to the
00:48launch of GPT 5.6, even though GPT 5.5 literally just released.
00:53And honestly, based off of some of the early testings I've personally gotten a chance to
00:58do, GPT 5.6 feels stronger than the upcoming Gemini 3.5 Pro, which is also expected to drop
01:05within the next month.
01:06Essentially, GPT 5.6 is shaping up to be a June release, and there are a lot of interesting
01:12things surrounding this launch.
01:14Meanwhile, with DeepSeek, remember the discount that was basically given to the DeepSeek version
01:204 Pro, the 75% pricing discount, well, that is now permanent, meaning you're getting near
01:27state-of-the-art performance at dramatically lower costs.
01:30There's a lot more happening in this space, and there's a lot to cover.
01:34So with that thought, let's just simply get into it all.
01:36If you want the best AI tools, workflows, and drops before everyone else, join my free
01:42newsletter with the link in the description below, which is completely free.
01:45Let's start things off with Mythos 1, Anthropic's new flagship model, or also known as Cloud
01:51Mythos 1 Preview, which is what the newly discovered Model Slug is currently displaying.
01:56This is a model that is appearing to be getting prepared for deployment across Cloud Code, Cloud
02:02Security, and several other Anthropic applications.
02:05This is the flagship model that was recently hyped up by Anthropic, and the model itself is
02:12briefly something that has become visible inside Cloud for a short period of time before disappearing
02:17again.
02:18But beyond that, additional backend strings are referencing Mythos as been added as well, including
02:24mentions of accessing the Cloud Mythos model inside Cloud Code, as well as a couple of different
02:30applications.
02:31That said, this still does not necessarily mean that the general public release will be coming
02:36fairly soon, or will be of this direct model.
02:39Anthropic previously hinted that some of these internal Mythos variants may be reserved for
02:46specific products, testing groups, as well as enterprise and security tooling, or future
02:51agentic infrastructure, rather than being exposed as a standard selectable public model.
02:56Regardless, I still believe that Mythos will be released, some sort of distilled version,
03:02probably within the next few months, latest by October.
03:06Now, Testing Catalog had done a great report on this leak.
03:10And what's interesting is that Anthropic now seems to be far more open to eventually releasing
03:15the Mythos publicly compared to earlier messages.
03:18And you can see that from their recent tweet where Project Glasswing has an update where they
03:23stated that Mythos class models could reach general release soon because of strong safeguards
03:29that are being placed right now.
03:30And that also hints why we're starting to see it within Google Cloud, as well as AWS's
03:37vulnerability discovery program, which is definitely suggesting broader infrastructure
03:41testing, which is already happening behind the scenes.
03:44So I'm not saying this could be released fairly soon, but these are usually hints that you would
03:50see when a model is basically almost close to deployment.
03:54Now, in parallel, Sonnet 4.8 and Cloud Opus 4.8 model selects were recently spotted on Google
04:02Vertex AI.
04:03This is usually a clear sign that both models could now actively be prepared for a release,
04:09whether that is in June or even late May, which could be in a couple of days.
04:15Now, rumors are also pointing towards select Anthropic partners already conducting internal
04:20evaluations behind the scenes.
04:22Now, a launch within the coming weeks would definitely line up closely with the cadence
04:27Anthropic has established following the Opus 4.7 back in April, while also fitting neatly
04:33alongside the broader Mythos and Cloud Security rollout, which is currently taking shape.
04:38But regardless, the Sonnet 4.8 leaks are also starting to paint a pretty interesting picture
04:43of where Anthropic is heading next, because apparently Anthropic accidentally shipped a massive 500k
04:50internal debugging source map through a Cloud Code NPM update back in March 31st.
04:56And inside that leak, there was references to Sonnet 4.8 already, which is hidden within
05:02unreleased keyword filtering systems that strongly suggested that Anthropic may be skipping Sonnet
05:074.7 entirely. From what currently is rumored, Sonnet 4.8 is expected to inherit some of the
05:15major version upgrades that were introduced to Opus 4.7, potentially pushing visual understanding
05:20accuracy above 98% for things like UI mockups, screenshots, and messy architecture diagrams.
05:28Coding performance is also rumored to be getting a pretty massive jump, especially around cleaner
05:33one-pass completions, and much more literal instruction following. There are even mentions
05:39of a new X-High reasoning efforts here, which could offer stronger reasoning capabilities without
05:45the extremely long generation time some higher reasoning modes currently suffer from.
05:50But one trade-off though that many have noticed from these leaks is that the updated tokenizer
05:55may reportedly consume around 30% more tokens on the same prompts compared to older Sonnet models.
06:02Right now, Sonnet 4.8 is definitely expected to be releasing mid to late June, and I am definitely
06:09excited for it. But still at the back of my head, I'm scared because it's an Anthropic model, and these
06:15guys have some sort of problem with rate limits, and due to the shortage of compute that they have
06:21right now, I'm not too sure and confident of any model draw from Anthropic right now. Now moving on to
06:28OpenAI, there are GPT 5.0 leaks that are starting to ramp up fast ahead of what's currently expected to
06:35be launching in June. OpenAI's researchers are recently hinting that the model is responsible
06:41for major math breakthroughs internally, which is already being used as a daily driver model for
06:47debugging as well as technical workflows, which is pretty interesting considering GPT 5.5 only just
06:54launched like late April. During deployment, internal testing tags like iris alpha as well as ember alpha
07:02and beacon alpha were also spotted, potentially pointing towards multiple GPT 5.6 variants being
07:09actively evaluated behind the scenes. And I've actually gotten a chance to actually use it throughout
07:14the arena, and I gotta say it is quite impressive. From current leaks, the GPT 5.6 appears heavily focused
07:21on stronger multi-step reasoning, it's better at agensic workflows, as well as significantly improved
07:27with its front-end generation quality. There are also canary rollout references that have also started
07:33appearing in developer environments, which mirrors the same quiet testing patterns that OpenAI usually
07:39uses before they end up launching a model. This is something that they've actually used for the GPT 5.5
07:46before they had launched it. Right now, leaks are pointing towards two separate releases.
07:51You have the GPT 5.6 as well as the 5.6 Pro, and honestly, it looks like June is
07:57going to be packed
07:58with many major releases because you have Sonnet 4.8 or Opus 4.8, whatever it may be. You also
08:03have
08:04Gemini 3.5 Pro, which is confirmed, and now most likely GPT 5.6. A new gift a week of
08:10co-work referral
08:11program where users are reportedly receiving three invite codes that they can share with others where
08:17they state that send a friend a free week of co-work, and if they love it and subscribe, you'll
08:23get 10 euros in usage credits, which is definitely a lot. But what's notable here is that Anthropic is
08:29also bundling access to cloud integrations around Excel, you have PowerPoint, Chrome, as well as cloud
08:36code as a part of the experience and showing that they're clearly pushing co-work as a much broader
08:42productivity ecosystem rather than just another AI chat app. Now, I believe this was there before,
08:48but this new co-work addition is fairly new. And lastly, DeepSeek just made one of the biggest
08:55pricing moves that we have ever seen in the AI space recently because the company officially
08:59confirmed that the massive DeepSeek version 4 Pro discount that they recently launched with the drop
09:05of DeepSeek version 4 is going to be permanent. This basically means that you're using DeepSeek
09:10version 4 Pro, getting nearly state-of-the-art performance on almost every generation, which is
09:17now going to be roughly 75% cheaper forever. This is listed at $0.43 per 1 million input tokens,
09:24and it's listed at $0.87 per 1 million output tokens. This is quickly turning DeepSeek version 4 Pro
09:31into the new king of cost per token economics for developers building large-scale agendistic
09:37systems and AI apps. I'm not saying DeepSeek is the best model, but this is something that you can
09:42definitely use as your daily driver to get most out of the model whenever you're working on hard,
09:48repetitive tasks. Now, just to put things into perspective, we have DeepSeek's pricing right here
09:54listed, and if we are to compare it to OpenAI, we can see that GPT 5.5, the flagship model,
10:01is listed at $5 per 1 million input tokens and $30 per 1 million output tokens. Now, if we take
10:07a look
10:07at what Anthropic is doing with the Opus 4.7, it is similar, $5 per 1 million input tokens and
10:13$25
10:14per 1 million output tokens. But if you are to compare this to DeepSeek, you can see that this model
10:20undercuts both of them massively while still competing in the same tier for many coding and
10:26reasoning workloads. That's honestly a pretty major shift for the AI market because it puts
10:32enormous pressure on Frontier Labs around price efficiency, especially for developers running
10:38large-scale agendistic workflows where token costs can stack up extremely fast. Usually, I'd end this
10:44segment off with some fun AI robot clip. But honestly, this is something I wanted to briefly talk about
10:50because the level of misinformation around AI lately is getting pretty wild and honestly
10:56unnecessary. I know a majority of you watching are already huge supporters of AI and probably
11:02understand what I'm about to say. But if I'm able to even slightly help shift the conversation
11:07towards what AI is actually doing for society instead of all the fear-driven narratives online,
11:13then I think that's important. There were even protests recently on Granville Bridge
11:19against AI data centers where people were chanting all sorts of things about AI. And look,
11:26conversations around energy usage, regulations, and job displacements are completely fair to have.
11:32But treating AI like it's inherently evil completely ignores the positive impact this technology
11:38is already having on people every single day. AI is helping accelerate scientific research,
11:44improving accessibility, assisting developers, helping small creators compete with massive companies,
11:51and dramatically increasing productivity for millions of people worldwide. A lot of the loudest
11:57reactions right now are honestly coming from fear, misunderstanding, or people only seeing viral
12:03doom post narratives online instead of actually using the tools themselves. Like any major technology
12:10shift, there are going to be risks that need to be managed responsibly, which is what we're trying to
12:14see with what Anthropic is doing with Mythos. But there's also an unbelievable amount of good that can
12:21come from this if it's developed properly. And that is why I tend to push AI out on a regular
12:26basis,
12:27because it can definitely change anyone's life.
12:30If you like this video and would love to support the channel, you can consider donating to my channel
12:35through the super thanks option below. Or you can consider joining our private discord, where you
12:40can access multiple subscriptions to different AI tools for free on a monthly basis, plus daily AI news
12:47and exclusive content, plus a lot more. Now I hope you enjoyed today's video and got some sort of value.
12:52I'll leave all the links that I used in today's video in the description below, but make sure you take
12:56a
12:56look at the second channel, join the newsletter, join the discord, follow me on Twitter. And lastly,
13:00make sure you guys subscribe, turn on the notification bell, like this video, and please
13:03take a look at our previous videos so that you can stay up to date with the latest AI news.
13:07But with
13:07that thought, guys, have an amazing day, spread positivity, and I'll see you guys really shortly.
13:11Peace out, fellas.
Comments