We're working towards the open intelligence safety stack! We've always been interested in making models do 'real' work. But we've also been very interested  in making sure they are safe and aligned (before it was cool). This started for us with mech interp at @parsedlabs, and now continues at @baseten which serves so much of the world's traffic. The responsibility definitely lies both on the closed and open labs to help shape this work. Excited to partner with @Thom_Wolf @ClementDelangue @julien_c + team from @huggingface and @eric_ho @banburismus_ + team from @GoodfireAI. We'll be publishing our safety research in the open and are aiming to build as much of it as possible into our serving stack. To start, this will include alignment training, monitoring in prod, and interventions shaped from these detected failures. A big ecosystem effort!

Safety is not just for closed models. The closed frontier labs are a canary in the coal mine for what is coming at scale. They give us a glimpse into the future and a window to harden our systems and prepare for abundant intelligence, with all the risks that come along with it. The OpenAI agent swarm attack on Hugging Face is the kind of failure we need to prepare for as open-source models catch up. The providers serving those models (such as Baseten) have a big role in establishing what safety and monitoring standards look like. We’re proud to be taking the lead on this at @baselabs with our collaborators @huggingface and @GoodfireAI. We’re developing safety research in the open and building it directly into Baseten’s inference infrastructure, with the aim of making it available to all our customers. This includes training models to follow explicit policies, detecting failures at runtime, and connecting those signals to controls that can intervene. We invite others in the open-source ecosystem to join us in building the tools and standards we’ll all need.

The future is about continuously turning your valuable data & ideas into better/faster/cheaper models -> Owning your Intelligence the Baseten team rocks at helping us run tons of experiments to make this happen @AEllisBloor @mudithj 🤝 A lot of work in our Labs team looks like: - sitting down with a buddy @jakebroekhuizen - picking a problem/agent to make better, cheaper, faster - curating data + evals/environments - doing tons of training experiments, with continuous measurement of every tweak - Loop (pun intended) the infrastructure and tooling is here to allow every team to run hundreds of experiments and get the best quality/cost/latency tradeoff for their customers and use-cases also ngl training is fun and your agents will love the experience

A flavour of our research interests on the Dwarkesh Podcast this week, discussed by our very own @oneill_c. Just the start of a longer conversation about long-horizon RL and frontier open-source training here at Base Labs.

I think one of the more interesting things we debated here is whether RSI is a cumulative task. Attention plus MoE plus GRPO etc seems to me like a line in the sand that you can just add to the stack once you discover it. You don't need to take five steps back to take 10 steps forward. But a lot of the work in the world isn't this clean and it certainly isn't this stationary eg legal work. This leads to some perhaps unintuitive predictions such as why RSI might land before continual learning (and why it's going to be hard to get off the current paradigm even if it's wrong) Thanks for having me @dwarkesh_sp!

Awesome work by @nikogrupen @ItsJulioPereyra and team with @baseten. Together with @baselabs, we built a recursive language model (RLM) harness that lets a root agent delegate document review to sub-agents and combine their findings into a diligence memo. Post-training Qwen 3.5 in this RLM harness more than doubled rubric pass rate on LAB Diligence tasks, from 29.9% to 63.0%, and increased review coverage from 62% to 96% of all documents in a dataroom. We’re also doing an RL scale up run with GLM-5.3 and will share the results soon and believe model-harness co-optimization is a viable path to automating end-to-end legal tasks like M&A diligence.

This is how we support a bunch of neolabs training frontier models (some of which we should be able to talk about in public soon when they release their models). Pretty neat trick @part_harry_ and @stefanopopoulos cooked up reading the compressed block directly rather than scanning every byte.

Excited about this RLM work as it shows: (i) Clear benefits of harness design that solves for the particular constraints of the task (every model showed significant uplift with an RLM harness). RLMs here solve for the 80M tokens sometimes required to be traversed per task (ii) RL works and the recipe can translate (with some careful monitoring). We RL'd smaller models like Qwen 122ba10b and larger ones like GLM 5.3 (iii) RL shapes useful behaviours even when not directly rewarded. In this case, dataroom coverage went from 62% to 96% which intuitively is the right behaviour being learned. We're just getting started etc. etc. @nikogrupen @gabepereyra @ItsJulioPereyra @winstonweinberg and broader research team are cranking.

Today we're launching Base Labs, a research lab by Baseten. Our mandate is to make open-source as useful as possible, and our one rule is that we publish without exception, including what fails. Up until pretty recently I thought the way to get the world onto open models was to train them for one company at a time. @mudithj, @maxkirkby and I cofounded @parsedlabs on that bet, @baseten acquired us, and we spent the last year running their training team doing it for customers one by one. Every one of those engagements taught us something new about how models learn, forget, specialise and get cheaper, and almost none of it got spoken about. Sadly, in general that's the field's default in that the people who know the most about training have the least freedom to say it. To be clear I don't think the closed labs are the villains here. They get to new capabilities first, which buys the rest of us time to harden the world before that stuff is everywhere, and they're the ones paying to find out what's actually possible. But I'm fairly convinced the only real advantage they have is data and scale, and their incentives point squarely at the frontier. You can't do slow, public science on how these things learn when your job is the best model by end of quarter. Someone without that pressure has to, and there are very few of those someones around. Hence Base Labs. We have the broad remit of making open-source as useful as possible and our one rule is that we publish without exception, including what fails. The first problem is continual learning, which I have come to think is several problems wearing one name. We are also working on the open RL environments and data that open models need and currently can't get, because we have to aggregate data with the same ferocity everyone's been talking about aggregating compute. Plus a bunch of other stuff I'm genuinely excited about, eg a safety stack people can run on top of open deployments, and performance research so these things are cheap for everyone to serve. I still think open and closed coexist, and that's the good world. It's just that coexistence isn't free, someone has to actually do the work, and this is basically what keeps me up at night. We're hiring researchers, engineers and fellows. Come help distribute the mandate of heaven!

Baseten Head of AI Model Training @oneill_c says the future is many specialized LLMs dedicated to specific tasks, with bigger labs deployed on the frontiers of areas like science and math: "People are thinking about intelligence capabilities in the wrong way. People are thinking about intelligence relativistically. They say, 'OK, the open-source gap is like 6 months behind closed-source, and GLM 5.3 is as good as Opus 4.8,' or whatever." "The best way to think about what models can do for you, and for the world, is in an absolute sense." "So for any given task that you want to do with an LLM, there's some intelligence threshold where below that you can't do the task, and above that you have very diminishing returns to more intelligence on the task." "So when you think about it that way, the game of LLMs over the last 5 years has been, 'OK, we have these things we want to do with them. Closed source hits it first... but open-source can eventually do that task. And then for many reasons, once you have the base level of intelligence required to do it, you probably do want to swap to open-source." "It's not really about the [frontier lab] God model being better. Like, if I'm filing a tax return, there is a limit to how much intelligence I need to do that particular thing." "So I think the world is going to look like — frontier closed-source labs are going to continue to push the frontier. You do want to use the most intelligent model. You have very inelastic demand for intelligence when you're doing frontier science or frontier math." "But for a lot of the economically valuable things, it looks a lot like, 'I'm a Cursor, or I'm one of these big companies who are realizing I can't just be a wrapper anymore. I've been through the life cycle of building a product that people love. And I should be using that information to make my model better at the things that I care about, and not at anything else.'"

Some questions demand the most accurate answers. @youdotcom built its Answer API to help users get grounded, cited answers to search queries in real time. We're proud to power the inference behind it, helping lower costs by 30% and match closed-source frontier model accuracy with open-weight models. Full story here:

Long horizon legal agents à la recursive language models coming soon. This is our best (current) solve for how to effectively process 10s of millions of tokens’ worth of context in the legal domain. Don't know many other vertical AI companies moving this fast, writing this openly, and at the frontier. Big props to @gabepereyra @winstonweinberg @nikogrupen @ItsJulioPereyra and team for pushing

At Baseten, we have a thesis: obsessives move the world forward. Few people embody that more completely than @bryan_johnson. On Wednesday, August 26th, we're hosting Bryan Johnson at our San Francisco office.

I really hope that in 6 months, when there is (yes) another cyclical swing towards closed source, we all remember the energy and values that are currently in the discourse. It’s very easy to rally around open source when the line between the two frontiers is this blurry. But the reason we’ve rallied behind this thesis for so long is that in order to build this safe and democratised future of intelligence, it should be shaped by many, red-teamed/interpreted by many, and served by many. Yes, I really do believe that the safest (and most fruitful) world with diffused AI is the one contributed to by more of the brilliant people in our field than the ones just behind closed doors. And, as so many have said (including @OpenAI), we want a world of open and closed source models that push humanity forward. Proud to be working directly with so many of the teams that are part of this.

Doesn’t this fire you up? We’re gonna do it. We’re actually gonna do it; open source is going to diffuse this intelligence to everyone. Frontier closed labs will still make money but they won’t control the light cone. The recipe is commoditised, now open source just needs to overcome the organisational and menu costs of aggregating data and compute. The good timeline is actually winning