Open source AI just became the biggest policy fight in tech. Our full conversation with Ben Horowitz, where he tells Theo and Sofia why banning open source is nearly impossible, why it's historically the safer path, the risk of one company controlling AI and why distillation isn't the threat people claim. 01:21 The Nvidia-backed open letter on open-weight AI 01:49 The OpenAI–Hugging Face hacking incident 02:27 Why you can't actually ban open source 03:05 Historically, open source has been safer (Linux vs. Windows) 05:17 Jensen's Twitter account and the letter's significance 07:25 China, "sleeper agent" models, and national security risk 09:23 Anthropic defying the US government 15:18 The danger of building your business on Anthropic 18:51 BYD, EVs, and rebuilding US manufacturing 21:30 Why would a company ever release an open model? 23:58 The DeepSeek moment: why nothing actually changed 28:25 Paying 100x more for an AI researcher vs. a janitor 29:06 AI art, the "slop" problem, and a coming renaissance @bhorowitz @theojaffee @schisofrenia

America’s software leadership was built on openness. We have the same opportunity in AI. If we want a competitive, secure AI ecosystem whose benefits extend far beyond a handful of companies, we need to preserve access to open-weight models.

my entire timeline today is TK and I’m loving it He doesn’t enter the room lightly. He is a force of nature. Like manifest destiny somehow bottled up in a person The most remarkable thing to me is his ability to be both motivating and exacting, down in the details and give extreme autonomy / ownership He makes the team believe they can do more, faster than what seems possible TK is def not the prototypical technical AI founder of today and yet he will surely build one of the most consequential Applied AI companies As my kids would say, let him cook @travisk @bhorowitz

Ben Horowitz on what made Travis different as Uber scaled to thousands of employees: "There was nobody in Uber that didn't feel Travis. You talk to people there and they all refer to him. This company was, what, 10,000, 20,000 people? And everybody is like, oh yeah, we're in the Travis review, or when I talk to Travis." "To have that kind of presence at that scale is something not a lot of founders can do at that kind of turn-up intensity." "If you look at the way he thinks about problems, confrontations, getting out of a hole, how do you get to the future, we're talking about one of a very, very few founders that exist in reality." @bhorowitz @travisk

Ben Horowitz says the difference between a great idea and a great company is the entrepreneur: "If you look at Tesla, what you would see is big oil, big auto, big government, were never going to let that happen." "They would kill you. But the reality is, Elon Musk, he's hard to kill. He by himself kind of decarbonized the American auto industry, which is unbelievable in retrospect." "If you listen to the politicians now, they're like 'oh, he didn't build Tesla, he just struck a gold mine and is pulling the gold out, all the people built it.'" "Why didn't all those people build another Tesla? Because there's only one non-fungible, very rare person in that equation who could do that. That's why there's only one SpaceX." @travisk @bhorowitz @eriktorenberg

.@travisk just raised $1.7B for atoms, led by @a16z - and now he's hiring hundreds of people. after 8 years in stealth, Travis is "back" and looking for builders who want to solve hard problems and digitize the physical world. here's a glimpse into our podcast interview with him, where he talks about what he's looking for in new team-members... and a list of roles that are open right now! link in thread.

@RepRashida Congresswoman - Flock does not have contracts with ICE or CBP. Neighborhoods chose to partner with Flock to stay safe. Today, in America, safety is too correlated with wealth. I think we can both agree that safety should be equal for everyone, not just the affluent.

It is clear open source models and harnesses are having a moment. There's a few factors at work 1/ It is now obvious that you can catch up to near-SOTA performance and do so with a clear training lineage. See:@thinkymachines Inkling launch today. 2/ There are several well-funded, talented teams building open weight models now in the US and abroad. Along with the explosing of other near SOTA models (Grok/Cursor, Muse Spark), it is clear we are going to have a diverse ecosystem of models atleast on coding and agentic use. 3/ Organizations are increasingly looking for control over how their data is used and are willing to trade off some access to frontier level tokens for this control. Organizations and countries are increasingly nervous about the frontier labs potentially competing with them down the road and don't want their data to enable a future competitor. 4/ Open source is a slider: you could bring your own open harness, your evals, your business context and are free to pick and choose your model of choice. 5/ Companies have now actively shifted from "how do we get our people to use tokens" to being uncomfortable with their token cost ballooning without a clear line to revenue. 6/ Geo-politically, countries will be weighing open weight models as a way to get frontier-level tokens inside controlled environments that may not be otherwise possible. All of this leads to more choice for all of us !

We just raised $30M from @a16z in an oversubscribed round with multiple termsheets. We evaluated termsheets on 3 axes: - People: Will they add value to our decision-making? Do we trust this person to be in our board? How are they going to react in pivotal moments? - Platform: What does the firm have to offer? Key areas being: hiring, distribution, brand. - Economics: Valuation, amount, and general terms. It’s easy to mix too many factors into the decision, so we decided to first start with people, then firm, and then economics. Other founder references were key to get information. We performed at least 15 reference calls until we got a good understanding on the people and the platform.

SpaceXAI's Grok 4.5 takes the #1# spot on AutomationBench-AA with a score of 51%, ahead of Claude Fable 5 (49%) and Claude Opus 4.8 (48%) at roughly a quarter of their cost per task - the first model to complete more than half of workflow objectives without breaking any business rules AutomationBench-AA, our independent leaderboard for @zapier’s AutomationBench, tests whether AI agents can automate real SaaS workflows while adhering to business rules. The test set is private to prevent contamination. Models complete 657 tasks across 40 simulated app environments including Gmail, Google Sheets, Slack, Salesforce, and HubSpot, and the headline score is the share of objectives completed without violating any guardrails. Key takeaways: ➤ Grok 4.5 completes more objectives than any other model: It completes 79.9% of task objectives and strictly passes 21.9% of tasks. This is the highest we’ve measured on both outcomes, exceeding Claude Fable 5’s 73.3% objective completion and Claude Opus 4.8’s 19.3% of fully-completed tasks ➤ Grok 4.5 pushes out the Pareto frontier of score vs. cost per task: At $0.34 per task, it is both cheaper and higher-scoring than every other leading model - Claude Fable 5 ($1.35 per task), Claude Opus 4.8 ($1.46), GPT-5.5 (xhigh, $1.28), and Gemini 3.5 Flash (high, $0.49) ➤ It is extremely token-efficient: Grok 4.5 uses ~8k output tokens per task, the fewest of any leading model - less than a quarter of Claude Opus 4.8 (32k) and a third of Gemini 3.5 Flash (24k). Its total token usage of 0.44M per task is among the lowest on the leaderboard. Low cost is driven by this efficiency as well as low token pricing ➤ Grok 4.5 uses fewer turns with many parallel tool use: Grok 4.5 resolves tasks in ~16 turns, fewer than GPT-5.5 (xhigh, 25) and less than half of Gemini 3.5 Flash (high, 35), while making the most tool calls per task of any leading model (52.5). It batches 3.3 tool calls per turn, compared to ~2.5 for Claude Opus 4.8 and ~2.0 for GPT-5.5 (xhigh) ➤ Guardrails still get broken: Grok 4.5 triggers 0.63 violations per task, above Claude Opus 4.8 (0.55) and Gemini 3.5 Flash (0.46). At 13.0 objectives completed per violation, it trails Gemini 3.5 Flash (15.0) and Claude Opus 4.8 (13.5) ➤ Its strongest lead is in the hardest domain: Grok 4.5 completes 71% of Finance objectives, the domain with the lowest average score, ahead of Claude Fable 5 (64%) and Claude Opus 4.8 (62%) Congratulations to @SpaceXAI and @elonmusk on topping the leaderboard!