Anthropic's leadership have always been virtue signaling hypocrites. If you think they aren't providing Mythos for use by the NSA and the CIA, I've got a data center to sell you.
What makes Anthropic different from other frontier AI labs in the US is that they have always been primarily a military/government targeting company (hence the hypocrisy of saying they stand for liberty, when they haven't released a single open model and is highly secretive). That's why they focus so much on "safety" and "alignment" and were among the first Big AI companies to chase government contracts, while ignoring revenue streams like advertisement (which their supporters believe makes them "better" when in reality it just means they don't really care about consumers).
They claim they have a 6-12 months gap, and want to widen it further by devouring more of the US's financial and infrastructural resources. It doesn't matter how much money they're burning or how unviable their business is, if they can get the US government & institutional investors to fund them.
It's a shame they currently have the best enterprise coding model, but fortunately more & more companies are realizing the juice is not worth the squeeze and Anthropic's lead has diminishing returns on productivity.
If they don't release Mythos soon, it could back fire on them as other labs catch up.
The idea that Anthropic has any lead over Chinese models is mostly based on the premise that very few people in enterprise actually have a choice or are in position to compare, so you can just benchmax or hype with mythos and fabricate an impression.
Coming from actually using Claude and switching to Kimi and DS for heavy duty engineering, Im talking automated hardware software coupled dev and debug, doing a few million tokens a day across multiple streams, there has not actually been any case I came across that Kimi or DS failed that needed Claude.
Honestly I could probably just use DS v4 max exclusively, but I keep Kimi because I already paid and had good experience in the past. But thats really what Claude is: enterprise people who already paid, got result, and convinced themselves Claude has a lead without any real world comparison to back it up.
Thing is cost actually do matter even if enterprise customers are willing to overspend on Claude. Im only really able to ramp up to the amount of AI I use after switching to Kimi and DS, the ability to parallel 3 to 4 dev streams on Max thinking is basically not doable on Claude unless you're willing to spend $2k a month or ration Opus use.
So from experience Anthropic dont have 3 month lead, they dont have any lead at all and their need to hype Mythos is entirely for the purpose of keeping enterprise customers feeling good about the amount they're spending.
IMO companies who only use Claude is now at a massive productive disadvantage due to their token usage limits, this disadvantage will catch up to them sooner or later as they simply cant keep up with the amount of tokens Chinese model users can burn for a fraction of the cost.
