> Note: "Benchmarks are less important than real-world tests for production adoption"

> Significantly better SWE-Bench (+56 pts), MCP tool use (2x), and agent workflows.

What? Make up your mind do the benchmarks matter or not?

You can still get "new" ones on amazon in europe.

7900XT has 20GB and you can still get some unused ones.

R9700 has 32GB and is cheaper than most NVidia consumer GPUs, even though it's a "pro".

And I can still buy a new W7800 48GB for a relatively decent price.

In other news, car completes marathon in under 30 minutes...

Not my experience, with a couple of 7900XT that is a 4 year old card by the way, you can run qwen 3.6 at home. Which for most intends and purposes is comparable to sonnet 4.6

My guess they are trying lower quantization. I have not noticed I recently begun trying qwen 3.5 locally.

> Could launch a swarm of 100s of drones.

As far as I know we have never seen that happen against a single target. I believe the reasons are operational not cost related. A single truck can fit like 5 shaheds. For 100 at the same target at the same time you need to coordinate 20 crews just to get them in the air all these drones need to be controlled to some degree as well. It's possible but we have not seen such an attack. We have seen hundreds of drones targeting hundreds of targets against an entire country. So it's definitely possible, but I wager it's harder than it sounds to send 100s of shaheds against a carrier strike group.

Shahed drones are very slow, and can thus be very easily distinguished from antiship missiles and can also be intercepted far befpre they reach the ships. You are thinking SM-2s. But the best way to deal with such a threat is a flight of f-18s with a bunch of laser guided rockets (like 50 or 70) and a targeting pod, intercepting the drones hundreds of miles from the target.

Have you seen all the Chinese light show drone videos?

See them fly in massive coordinated swarms with precision?

See them automatically land in charging docks in waves by the thousand?

Those videos are not showing the world just a pretty light show.

It's the radars really for destroyers. The bridge is not actually where the ship is run during combat.

There is a room called the combat information center, that's where the ship is run from during combat, and that is behind armor, even in modern warships.

Additionally ships are separated into semi independent zones, that can take control of the ship, and continue fighting even if the rest of the ship is on fire.

The real liabilities are the radars, and the rest of the sensors in surface combat ships and the airplanes on deck in the case of aircraft carriers. Aircraft carriers in general are heavily armored compared to other modern warships and it takes a significant amount of firepower to even disable them much less sink them.

It proved nearly impossible to sink the Bismarck and Yamato battleships in WW2 just by shelling them.

Both were rendered useless hulks long before they went under, though.

Considering how the sunk ships at Pearl Harbor were refloated, refitted, and put back into service suggests otherwise.

That's great if you're in a shallow anchorage (average depth: 45 feet). Less so if you sink in the Arabian sea and you're under fire during the refloating process.

I also suspect modern ships are a little more sensitive to complete immersion.

Case in point: https://en.wikipedia.org/wiki/Helge_Ingstad_collision

> In May 2019, the Minister of Defense was presented with a report from Defense Material which concluded that a possible repair would cost 12–14 billion and take more than five years. The cost of purchasing a new corresponding vessel was estimated at NOK 11–13 billion, with a completion time of just over five years.

And it didn't even go all the way under.

Limited range? Shaheds have over 2000 kilometers more than tomahawks.

And btw, if you can get a submarince close to your target, torpedoes and missiles are going to be much more effective than drones.

Space is limited on platforms, a submarine might have space for 60 drones or 30 missiles, given the immense cost of the submarine, going with the missiles is the right call.

The trucks launching shaheds that iran is using can fit like 5 such drones, a similar truck could probably fit 2 to 4 cruise missiles the only reason they are using drones is the rapid production and cost associated with drones instead of the cruise missiles.