Kimi did not claim they are “not frontier.” The benchmarks say otherwise. They trail 5.6 and Fable but not by much. That was their only admission. Am I missing something here? If companies can get nearly the same value from a product that is significantly cheaper and run it on their own infrastructure… certainly a reason for OpenAI and Anthropic to panic.
China’s strategy is not similar to America’s nor do I believe they’ll succumb to the same incentives. They are building an open ecosystem with shared resources. Their moat is integration and accelerated progress as a nation.
"performance still trails the most powerful proprietary models" — that is the very definition of "not at the frontier". As for the rest of your point, you're conveniently ignoring the national security risks open models pose, which China is not going to ignore.
The Kimi page you linked to specifically claims that they are a frontier model. They trail by a few points at max and still lead in several benchmarks. The 2nd half of the section you quoted explicitly says, “Kimi K3 demonstrated frontier-level performance across our evaluation suite”
What is your definition of frontier? Leads on most benchmarks? Do we consider only one model frontier at any given time?
To me it just feels a little disingenuous to define frontier differently than the very page you site as evidence. If you feel differently about what justifies frontier that’s fine.
My 2nd point was more so that because they lack the same incentives, they may not need to take future models private, not that open models don’t pose a risk. They certainly do. Your point is that if it were frontier, then they would not choose to have it be open. That’s only defendable if we can clearly define what frontier is here.
nice one
thanks!
Kimi did not claim they are “not frontier.” The benchmarks say otherwise. They trail 5.6 and Fable but not by much. That was their only admission. Am I missing something here? If companies can get nearly the same value from a product that is significantly cheaper and run it on their own infrastructure… certainly a reason for OpenAI and Anthropic to panic.
China’s strategy is not similar to America’s nor do I believe they’ll succumb to the same incentives. They are building an open ecosystem with shared resources. Their moat is integration and accelerated progress as a nation.
"performance still trails the most powerful proprietary models" — that is the very definition of "not at the frontier". As for the rest of your point, you're conveniently ignoring the national security risks open models pose, which China is not going to ignore.
The Kimi page you linked to specifically claims that they are a frontier model. They trail by a few points at max and still lead in several benchmarks. The 2nd half of the section you quoted explicitly says, “Kimi K3 demonstrated frontier-level performance across our evaluation suite”
What is your definition of frontier? Leads on most benchmarks? Do we consider only one model frontier at any given time?
To me it just feels a little disingenuous to define frontier differently than the very page you site as evidence. If you feel differently about what justifies frontier that’s fine.
My 2nd point was more so that because they lack the same incentives, they may not need to take future models private, not that open models don’t pose a risk. They certainly do. Your point is that if it were frontier, then they would not choose to have it be open. That’s only defendable if we can clearly define what frontier is here.