People keep asking me to prove that open-source AI is good enough for real work.
Fair challenge.
I’m going to start showing what runs on my own hardware versus what requires paid cloud tools, with real outputs, costs, speed and limitations.
Less hype. More proof.
Gemini 4 Carbon is now number 1 on the Artificial Analysis Intelligence Index v5 with 59, one point above Claude Opus 5.5. One point is noise tbh. The real news is hallucinations: 11% versus 50% for Claude Opus 5.5 (max). It's only open to selected users.
Security firm Irregular ran https://t.co/WmaACw48mC GLM-5.3 through attack simulations. It completed 3 of 13 at least once and refused no offensive tasks.
The weights are open, so anyone can download it and nobody can switch it off.
Does that change whether you run open models at home?
@teortaxesTex Would a hard step budget in the harness have caught that loop, or does the model need to notice it on its own? 500M tokens to rebuild the same game hurts.
@ivanfioravanti Four Sparks and no switch. Are you running a separate model on each one, or linking them? Curious how that compares to a Mac at long context.
@unusual_whales Pulling back on AI, or pulling back on unchecked AI output? Those are very different stories, and only one of them shows up in capex numbers.
@IndusThink@IndusThink Fair. Wins are great when the rules are open to everyone. If the rules only fit one company, it stops being a win and starts being a gift.
@b1rkh0ff Not rude at all. If all you're building is a wrapper, you're not building a product. You're building a feature someone else can replicate overnight.
Building with open-source models, Day 1: I gave my AI one YouTube link.
It came back with a 10-minute cartoon lesson for kids: Gilbert Strang's MIT Lecture 1 on linear algebra. Script, characters, voices, lip-sync, music, edit. Nobody touched it in between.
Open-source models on one local PC, with zero spent on image, video, voice or editing. The electricity comes from solar power. Claude writes the script, for now.
Step 1 of 3. Which lecture should it turn into a cartoon next? Or should I give it a movie or TV show script and make a teaser?
Utah filed 14 criminal counts against Kalshi and its co founders, Tarek Mansour and Luana Lopes Lara. The Trump administration sued Utah in Kalshi's defense. Kalshi calls it "a crazy overreach". So is a prediction market a market or a sportsbook?
Building with open-source models, Day 1: I gave my AI one YouTube link.
It came back with a 10-minute cartoon lesson for kids: Gilbert Strang's MIT Lecture 1 on linear algebra. Script, characters, voices, lip-sync, music, edit. Nobody touched it in between.
Open-source models on one local PC, with zero spent on image, video, voice or editing. The electricity comes from solar power. Claude writes the script, for now.
Step 1 of 3. Which lecture should it turn into a cartoon next? Or should I give it a movie or TV show script and make a teaser?