I went a dinner a few weeks ago with a bunch of enterprise execs who told me "we will never use Chinese models." "Even if it's 100x cheaper?" "No, we care about safety and security."
1. They don't understand when they host open-source models with their own GPUs or US data centers, they won't share their data to China.
2. They are giving away all their data to OpenAI and Anthropic rather than owning it privately themselves.
3. They don't understand math. 100x is a big number and lots of profits.
It's almost July 2026 now. If your execs still talk like that, fire them now.
we have been building at the frontier of making open source models accessible in a private, verifyable and compounding way that is completely self hostable in your own infrastructure!
today - we just went live as the first agentic AI workspace to server glm 5.2 in a fully confidential mode!
you can read more here: https://t.co/dX4MAHzZHB
Sovereignty in AI means owning the means that produce compounding value for your organization.
The journey starts with products you can verify and trust on your own.
Fluso is now integrated with the Prem Enclave API, and the full sovereign loop is finally closed.
one of the strongest models on the market fell off the face of the earth for three weeks.
if your database vendor did that, you’d have a DRP. what’s your DRP for the intelligence layer 👇
So basically Alex Karp’s argument is that frontier AI labs profit three times: (1) they charge you for tokens, (2) they get access to your IP and business know-how, and (3) they eventually commoditize your competitive advantage. Instead, he says enterprises should pay Palantir to deploy open models and keep their own alpha
LOL I did not expect him to suddenly become an ally for open source even though it is of course very self-interested
got a stern talking-to from my boss about "building a presence." so: fluso is a workspace where your AI agents run on your own infra instead of phoning home to openai. turns out banks and law firms love it when their data never leaves the building. ask me anything
Anthropic released Sonnet 5 this week. it scored a meaningful step up from Sonnet 4.6 on CursorBench, from 49% to 57%. On the same benchmark, @cursor_ai's Composer 2.5 is beating Sonnet 5, at a fraction of the cost per task and a fraction of the tokens.
Composer 2.5 is reportedly built on Kimi K2.5, an open-weight base model. Cursor has been continuously fine-tuning it on real usage data from inside their own product. when a user accepts, rewrites, or discards a code suggestion, that feedback becomes a signal that shapes the next iteration of the model.
Anthropic ships some of the most capable base models in the world, but they cannot see what code suggestions Cursor's users actually use or reject. Cursor can, and that access is what tipped the result.
Base models are becoming a commodity input to enterprise AI. compounding value now depends on whether a company owns the loop between the work its people do and the model that runs on top of it.
At @premai_io we are building this loop as a product, so you do not need massive engineering budgets to own it!
Curious who else you are seeing run this same play!
turns out banks and law firms don't want a smarter chatbot. they want the loop: an open-weight model inside their own walls, context compounding on data that never leaves. cursor built it by hand. we're making it a product
Anthropic released Sonnet 5 this week. it scored a meaningful step up from Sonnet 4.6 on CursorBench, from 49% to 57%. On the same benchmark, @cursor_ai's Composer 2.5 is beating Sonnet 5, at a fraction of the cost per task and a fraction of the tokens.
Composer 2.5 is reportedly built on Kimi K2.5, an open-weight base model. Cursor has been continuously fine-tuning it on real usage data from inside their own product. when a user accepts, rewrites, or discards a code suggestion, that feedback becomes a signal that shapes the next iteration of the model.
Anthropic ships some of the most capable base models in the world, but they cannot see what code suggestions Cursor's users actually use or reject. Cursor can, and that access is what tipped the result.
Base models are becoming a commodity input to enterprise AI. compounding value now depends on whether a company owns the loop between the work its people do and the model that runs on top of it.
At @premai_io we are building this loop as a product, so you do not need massive engineering budgets to own it!
Curious who else you are seeing run this same play!