Subquadratic's claims about cost are harder to verify because SubQ is not yet widely available. According to Dangel, it costs $2,600 to run Anthropic's LLM Opus 4.6 through RULER 128, a test developed by Nvidia to assess a model's ability to retrieve information from large data sets. And SubQ? "It cost us eight dollars," he says.
If this is legitimate their technique will eventually be replicated by open source LLMs and the massive data center requirement will vanish. Programmers will run private LLMs on their laptops that match what current, massive data center LLMs do.
> Subquadratic's claims about cost are harder to verify because SubQ is not yet widely available. According to Dangel, it costs $2,600 to run Anthropic's LLM Opus 4.6 through RULER 128, a test developed by Nvidia to assess a model's ability to retrieve information from large data sets. And SubQ? "It cost us eight dollars," he says.
If this is legitimate their technique will eventually be replicated by open source LLMs and the massive data center requirement will vanish. Programmers will run private LLMs on their laptops that match what current, massive data center LLMs do.
It's a big "if" but it would also probably blow up the hardware market too. No more ram modules that should cost $60 costing $350.
It's a big "if" but it would also probably blow up the hardware market too. No more ram modules that should cost $60 costing $350.