BasedAI @BasedAI_co

If a model you don’t control sits in your critical path, your roadmap has a silent co-author, your unit economics have a floor you don’t set, and your access rests on conditions you can’t see. Open weights are the exit. Your data stays home, your access can’t be revoked, and the cost gap is structural, not promotional. New on the blog: Own Your Stack. Don’t Rent It. basedai.co/blog/own-your-…

As a an open-source AI advocate, the only surprising part about the Fable shutdown was that something like this hasn't happened sooner. We're building @BasedAI_co with the strong conviction that businesses need sovereign access to AI they control. Now the market is starting to get it too. 👇

The first major closed beta update is live. The whole meeting sequence, handled by your AI coworker: Prep -> transcript -> summary -> action items done ✅ What we learned and what shipped: hirebase.co/blog/beta-upda…

A closed frontier model is a service. An open-weight model is an artifact. A service can be deprecated, repriced, fenced off, and, as proved when Fable 5 went dark for every customer at once, switched off entirely. An artifact can't. Once released, it's infrastructure: served by anyone, anywhere, indefinitely. The capability gap has mostly closed. What remains is a difference in who holds the keys.

"Open models are catching up" is over. They caught up. The question stopped being which model is best and became who controls the one you depend on.

Arena.ai @arena

2 days ago

Exciting news: GLM-5.2 (Max) ranks #2 in Code Arena: Frontend, with +29pt over Claude Opus 4.7 (Thinking) and only behind Fable 5! GLM-5.2 is the best open model vs Kimi-K2.6 and Minimax-M3 by a large margin. - #2 React and #4 HTML sub-leaderboards - Ranks as the top model in

158 491 4K 1.3M 1K

Renting a capability and owning an asset are not versions of the same arrangement — one can be taken back, the other compounds. Our CEO @TeanaTaylor takes a step back to consider the concerns that Friday forced into the open: not which model is best, but who controls the layer your business now depends on.

Teana Baker-Taylor @TeanaTaylor

2 days ago

x.com/i/article/2066…

0 2 11 1K 1

On Friday, a government directive forced Anthropic to switch off Fable and Mythos for every customer overnight. No tier, no region, no warning. Build a product on a model on Thursday, have no model on Friday. The decision was made several layers above your contract. A stack you rent can be repriced, rewritten, or revoked — and none of it requires your consent. The question is no longer which model is best. It is who controls the layer your business depends on. Open weights are the only answer where the answer is: you

A frontier model went dark today, not just for the lab but for every business on it. Build on a model you don't control and your uptime is someone else's permission, gone by an order you weren't in. Open weights remove that single point of failure. This is why BasedAI exists.

This week was about one number - not what a model costs, but what the work costs. @GetHirebased published a great article about it hirebase.co/blog/what-an-a…

Hirebase @GetHirebased

a week ago

A per-token price tells you nothing about what the work costs. So we priced three real AI jobs - a landing page, a GTM plan, daily inbox and calendar triage. New on the blog, with the receipts 👇 hirebase.co/blog/what-an-a…

1 0 4 804 0

@GetHirebased Powered by BasedAI

Friday signal Most teams run every AI task through one expensive model - the cold email at the same rate as the hardest problem they have. That's the overpay nobody put on the invoice. Get the results, without having to think on the complexities. hirebase.co

x.com/i/article/2065…

This is the moment open weights stop being the cheaper alternative and become the floor everything else prices against. The race used to be access. The race now is cost per outcome. The long version is on basedai.co

Depending on the model you benchmark and how cache-heavy the workload is, the same task lands 6 to 30 times cheaper on open weights. The capability gap closed last year. What's left is a price gap - and price gaps grounded in cost structure don't close from the expensive side.

A sale has an end date. On May 31, DeepSeek made its 75% V4-Pro price cut permanent - published rate from $1.74 / $3.48 per million tokens to $0.435 / $0.87, no expiry. A permanent price is a vendor pricing at the level its cost structure supports.

Most companies use AI like a marginally better photocopier. The ones who redesign their work around it see 5x the impact, and we've known why since 1881, when factories made the exact same mistake with electricity. The full story, new on the blog: basedai.co/blog/the-1-tri…

New Myth just dropped: "the model is included in your plan". Legend says if you check your subscription at midnight on the 22nd, it vanishes. Your Hirebase coworkers? Always running on the best-performing model for the task at hand. No countdown timers.

Most teams picked one expensive model and run everything through it - the cold email and the genius-level problem - paying the same rate. @JoshGoodbody on why you're paying a frontier price for work a cheap model finishes just as well, and what changes when the routing is somebody else's job.

Josh Goodbody @JoshGoodbody

a week ago

x.com/i/article/2064…

0 2 5 887 0

$1 of open weight work does what costs $30 on a frontier lab model. Same agentic task, same kind of API call, open weights underneath.