The internet consciousness has moved on from Anthropic's recent announcement of Mythos and their Project Glass Wing - but the industry has not.
I find it both interesting and, I can't help but phrase it as, a little concerning given the recent developments we've seen since I last posted about it. Around that time, the internet shifted its focus to how Claude had gotten lazy, would often give up on requests, or gave worse answers than it used to.
While this became widely accepted online - and I had colleagues mirroring this notion - it wasn't something I particularly noticed myself. It might have something to do with the way I personally use AI in helping me understand something complex when a quick Google search fails, polishing an important email, or writing an Excel expression in an unfamiliar domain. My use cases are perhaps more forgiving.
Regardless, Claude had gotten lazy. This was then followed by the release of Claude Opus 4.7 and (as we've come to expect at this point) quickly followed by a new version of OpenAI's Codex. Cue more drama and discourse on how it wasn't as good, over-explained things, or was simply a re-release of Opus 4.6 before its intelligence was throttled down - some of which should be inherently impossible while the other is true.
While the internet's focus was on Claude's performance - and how it stacked up against its closest competitors - the industry's attention was elsewhere and laser-focused on the promised Mythos.
Amazon was quick to follow with an initial investment of $5 billion and a soft promise of up to $20 billion more. More recently, it was announced that Google would follow in Amazon's footsteps with an investment of $10 billion, with up to $40 billion in total.
Wait, what? Google. Yes, Google. The company competing with Anthropic for the title of AI lab with the smartest model. The company with the best integrations, best AI image generator, and best model for robotics. That company invested in what may be their biggest competitor - certainly their biggest competitor in the AI space.
What did Anthropic show Google? This is the question I keep returning to.
Project Glass Wing was launched because Anthropic created something they were scared of. Something that proved difficult to contain. Something capable of finding security flaws with ease. A host of organizations agreed this was serious, and Google just doubled down on that assessment. Google - I still find this mind-boggling - Google.
The subtext is hard to ignore - when your fiercest competitor writes you a $10 billion check, they've seen something that transcends competitive interest. They've seen something that makes collaboration non-optional.
On a lighter note, DeepSeek V4 released the other day. This project has given me great joy since I first became acquainted with it, and it continues to do so. For all its flaws in avoiding prompt injections and more, DeepSeek represents to me what the AI industry could have been - open source and cost-effective.
While the capability isn't 1:1, the pricing comparison is striking. Opus 4.5+ runs at $5 in / $25 out per million tokens. DeepSeek V4-Pro runs at $1.74 in / $3.48 out. Combined with my view of how capable current models already are, and the biggest outliers are the integrations and harnesses used, this is a no-brainer. Unless you're expecting it to work with Claude Code, or are doing highly complex coding tasks, DeepSeek wins on API pricing alone. With their documentation and contributions to open source research, I struggle to hide my fondness of this project.
It's a reminder that not everything has to consolidate. That alternatives exist. But I'm not sure that's the direction things are heading.
Here's the bigger picture, as I see it. Open-source alternatives aside, we aren't just watching an A-team form, as I have previously speculated. The elite AI labs aren't just nearing the same intersection - they're melting together. Collaborating. Becoming one.
Given the ambitions of Project Glass Wing, one can hope this collaboration becomes what OpenAI was supposed to be at its founding. OpenAI - the company Dario Amodei left Google to help steer AI in the right direction. The company he later departed to found Anthropic with his sister Daniela, when its motives changed.
Dario Amodei's constantly shifting timeline for when AI will replace the need for human workers gives me pause. And I've perceived certain actions that Anthropic has taken to feel like a drift from their founding ethics. But Project Glass Wing, and the collaboration it's attracting, gives my more optimistic side reason to hope - perhaps Anthropic hasn't swayed as far from that original vision as I feared. And their continuous human approach to research with Claude strikes me as empathetic.
Although putting anyone on a pedestal usually turns out poorly, that same optimism tells me Dario Amodei might be the guiding force for good here - the person drawing the red line when it matters.
I work with data systems every day. The tools I'm integrating into workflows, helping others understand, advocating for where appropriate - these are the very systems whose creators are now pooling resources because the capabilities are outpacing individual oversight.
That's not something I can observe from a distance. It's something I have to think about in how I do my work.
The question isn't whether AI will reshape how we work - that's already happening. The question is whether the people building it are building guardrails fast enough. Glass Wing suggests they're trying. Google's check suggests it might even be working.
I find myself cautiously hopeful.