Amir Haghighat retweeted
Scaling past a trillion parameters requires inference engineers to solve 3 issues:
- Width: Routing between hundreds of sparse experts
- Depth: Mixing global and local linear attention
- Length: Processing 1M-token sequences
I asked Ali and Bryce to draw it up. They delivered.
33
171
1,773
129,746
Amir Haghighat retweeted
Baseten 🤝 Runway
We’re excited to announce @runwayml as a Model Labs partner, and to bring their video generation and editing models to Baseten Dedicated Inference!
- Gen-4.5 generates video from text and images, and
- Aleph 2.0 edits existing footage using text instructions and keyframes.
Learn more on our blog: baseten.co/blog/deploy-runwa…
1
1
47
2,525
We’re excited to announce @runwayml as a Model Labs partner, and to bring their video generation and editing models to Baseten Dedicated Inference!
- Gen-4.5 generates video from text and images, and
- Aleph 2.0 edits existing footage using text instructions and keyframes.
Learn more on our blog: baseten.co/blog/deploy-runwa…
3
5
65
6,703
Open models can come from anywhere. Securing them shouldn’t depend on where you get them.
Today we're launching Project Beacon to make safety a default for open intelligence.
We're partnering with @GoodfireAI to bring inference and in-line monitoring together for the first time, so safety signals are caught proactively.
Read more in our announcement: baseten.co/blog/safe-open-mo…
7
18
125
25,681
Our first research summit, All-Gather, is on November 18 at the Exploratorium in SF. Apply and see you there!
Progress that moves the frontier is almost never made alone.
We’re thrilled to announce All-Gather, our first-ever research conference that will bring together the brightest minds in research, founders, and engineers to join us for a day of debate and exchange.
Come hear Noam Brown, Sarah Guo, Horace He, Bryan Catanzaro, Jerry Tworek, Sabri Eyuboglu, Tuhin Srivastava, and many more debate pre-training, post-training, infrastructure, model performance, and emerging fields. Join us on November 18th in San Francisco.
Research has been part of our DNA from our early days of optimizing open-source models through our recent launch of Base Labs. We’ve grown with the open ecosystem, and we’re excited to keep building alongside it.
Space is limited, but we’d love you to join us. Apply here: baseten.co/all-gather/
1
2
54
2,388
Each week we share ongoing work from across the Base Labs team. We call these releases 'rollouts'.
This week we share our progress in understanding the low-dimensional structure of RL training and whether recent discoveries about rank-1 extrapolation of early updates really hold.
We find that performance gains largely come from easier problems and fixing doom loops, while harder environments are less linearly extrapolatable.
4
10
143
19,441
Amir Haghighat retweeted
Exclusive: Inside @baseten
The $13B startup at the centre of the biggest fight in AI - Open vs Closed.
We spent a day at their SF HQ with co-founder & CEO Tuhin Srivastava, Charlie O'Neill, Mudith Jayasekara and the Base Labs team - to understand why they're building an open ecosystem.
Backed by top VCs led by @saranormous @altcap @apoorv03 @wmccarthy11 @saammotamedi @nikiscevak @lachygroom @willreed + more. Plus participation from @nvidia.
0:00 The Open vs Closed AI War
2:18 A Billion AI Requests a Day
3:43 The Best Models Are Chinese
4:06 Tuhin: Don't Rent Intelligence
5:09 Sachin & Tuhin Talk Cricket
6:02 Tuhin: The Real Bottleneck
10:15 Why Launch Base Labs?
11:35 "You Have to Have an Opinion"
12:52 Centralised Control Is the Danger
13:07 Publish Everything, Even Fails
15:08 Hugging Face & the 6-Month Gap
16:10 Why We Still Need Closed
17
31
300
107,578
Two months ago I said the easiest way to use open models in Codex was `brew install baseten-switch`. Starting today you don't need baseten-switch: Baseten is the first inference provider in the new OpenAI B2B Marketplace. Enterprise OpenAI customers can run open models on Baseten natively via Codex and the Responses API.
baseten.co/blog/baseten-open…
8
13
159
1,643,968
Our second announcement of the day: Carbon is our newest sandbox infrastructure, built with more capabilities for long-running agents.
It has microVM isolation, an IPv6 address per sandbox, and snapshots and forks as first-class features. A sandbox snapshot can go to production within milliseconds; no extra build step needed.
Carbon is in private preview; DM me if you want access.
9
15
155
20,461
Amir Haghighat retweeted
Couldn't be more psyched to join the cracked team at @baseten to build the future of AI and help companies own the stack around their intelligence.
We're in our inference era 🌟
We're officially in our Sheila era. Excited to welcome our new CMO to Baseten!
baseten.co/blog/sheila-vashe…
23
5
121
14,837
Open models on Baseten can now search the web, within the inference path, thanks to our partnership with Exa, Keenable, Parallel, and You.com. A lot more to come here soon.
Until now, adding web search to open-source models meant hand-wiring orchestration, managing separate keys, and paying latency taxes on every round trip.
Today, we're solving that. We're excited to introduce Baseten Hosted Tools and Baseten Grounded Inference to bring real-time web search server-side to open models running on Baseten through a single configuration:
- 15% lower latency compared to client-side execution
- No extra vendor key required
- Zero orchestration
We're launching this preview version with four leading web search partners, @ExaAILabs, @KeenableAI, @p0, and @youdotcom.
Get frontier-level web search parity for your open-weight models. More here: baseten.co/blog/introducing-…
4
3
49
3,701
Amir Haghighat retweeted
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!
New episode with @johnschulman2, @oneill_c and @BerenMillidge.
I got together with some of the most insightful AI researchers I know who are at the openish companies, because I wanted to hear the details of what's actually happening at the frontier and what comes next.
0:00:00 – Steelmanning the case against RSI
0:18:39 – What’s driving the Chinese labs’ progress
0:28:06 – How will automated AI researchers be trained
0:33:51 – Will long-horizon RL elicit AGI?
0:45:24 – The sim-to-real gap
1:00:33 – How much progress is explained by data?
1:18:03 – Why is RL working so well?
1:24:54 – Move 37 and entropy collapse
1:28:31 – Rapid-fire timelines
11
15
160
28,197
“When will Baseten offer sandboxes” has been a common customer question. And we’ve taken a huge step in answering it: Baseten has acquired Blaxel.
The nature of the median inference call has changed in the past year. Models stopped just answering questions; they act now. A coding agent writes a patch, runs the test suite, reads the failure, and tries again, dozens of times to close a single ticket. Every one of those loops needs somewhere to execute code. And the agent itself needs its own runtime. And so do RL rollouts. The list goes on.
Running some code in a container isn’t hard. The hard problem is storage (state), networking (milliseconds-long cold starts), and security (VMs). That's what Blaxel has built: stateful, pause/resumable, VM-isolated sandboxes that scale to thousands of instances in seconds.
Zooming out, tomorrow’s cloud isn’t one of inference, training, or sandboxes. It’s one where these 3 primitives come together cohesively to meet users where they are and to unlock value that’s greater than the sum of the parts. That’s been the north star that Baseten and Blaxel share (they were in fact going to build inference and training). I couldn’t be more excited to do it together with Paul Sinai and team. Welcome 💚
baseten.co/blog/blaxel-is-jo…
8
17
117
16,639
AI labs run their managed RL rollouts on Baseten. Here’s how the team got to sub 40 second weight sync across global clusters.
DM me or @mudithj if you're interested in learning more.
1
40
3,693
Amir Haghighat retweeted
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.
9
9
94
12,867
Notion's AI Meeting Notes run on Baseten, and Baseten's knowledge base runs on Notion. Our teams have been working together closely to push the frontier on both fronts.
For AI Meeting Notes, our voice engineers built the fastest, highest-quality, yet lowest-cost solution on the market for speaker-attributed transcripts. We're proud to partner with the @NotionHQ team.
13
11
119
36,845
We have clearly benefited a ton from the ecosystem of ever-improving open models and the published research behind them. We will contribute back in a major way.
Today we're announcing Base Labs, a dedicated research organization focused on advancing open-source AI.
We believe in a healthy, open frontier model ecosystem. To enable this, we are working on:
- Blue-sky research on continual learning, the science of RL, and how models learn across their full lifecycle, with every experiment and recipe shared openly.
- The BaseHub Data Foundry: the highest-quality open RL environments, training data, and real-world benchmarks, built for anyone to train and benchmark on.
- Post-post training: taking open-source models and making them better, safer, and more aligned through continual post-training, built on our research, and deployed with our frontier safety stack so organizations can use open-source models with confidence.
- Making models cheaper and more performant through our model performance research.
This is a mission-driven research effort, not a commercial product. We believe the health of the open-source AI ecosystem matters and that the best way to advance it is to do science in the open.
We’re hiring engineers, researchers, and research fellows to advance this mission.
labs.baseten.co/
5
5
66
5,003
Even hackernews is raving about this model!
GLM-5.3 is live on Baseten Model APIs, day 0.
- The smartest open-weight model at 743B params
- 1M token context
- US only
- ZDR
Try it here: baseten.co/library/glm-53/
1
1
37
3,351
We're proud to be the fastest inference provider on Artificial Analysis, OpenRouter, and Hugging Face for GLM-5.3-Flash, at 122+ TPS. All served from the US only, starting on day 0, with ZDR by default.
Stay tuned for updates as our engineers continue to optimize GLM-5.3-Flash for even higher throughput and lower latency.
14
9
216
23,223