There's a post in Hot claiming that Algorand can do 2800 DEX TPS. But OP did not mention the misleading tactics and schemes that were used to make that claim

If you only want the TL;DR: the benchmark uses a burst of BATCHED instructions, counted them as TPS and compared it with other networks that did not do the same trick when in reality an average of only 28 signed TPS was recorded on chain in those blocks

Having 100 batched instructions in a transaction and claiming it's the equivalent of 100 TPS is completely misleading and false.

First of all, every blockchain in that list would gain a TPS boost if the scalability tests would have been done in the same way with batched txs, not to mention that it's completely useless to do that since real usage never behaves like that, you can't just batch 100 people's swaps under the same signature.

Second of all, the benchmark is missing thousands of user signatures that add significant strain on bandwidth and computation/verification making this claim utter nonsense.