Research, infrastructure, and data for building autonomous researchers.
Joined August 2025
- Tweets26
- Following2
- Followers5.2K
- Likes68
Kimi K3.1 Pro is now the best open-weight model at training other AI models.
But it’s still far behind Opus 5.5 at every budget.
New FrontierML results, our benchmark for end-to-end ML work ⬇️
Also new on the leaderboard: Gemini 4 Argon, DeepSeek V4.1 Flash (1017), and MiniMax M3.1 Flash.
DeepSeek and MiniMax are tied, and both are behind Kimi K3.1 Pro.
Gemini 4 Argon is on par with GPT-6.1 Sol, but behind the frontier.
We’ll keep adding models to FrontierML as they come out. If you want us to benchmark your model, reach out!
FrontierML is open-source: github.com/induction-labs/fr…
Leaderboard and full results: frontierml.org/
FrontierML has 21 realistic, end-to-end tasks inspired by research at frontier labs, neolabs, and academia.
Tasks are fully open-ended: agents can pay for compute, generate synthetic data, run experiments, use any public model or dataset, or design architectures from scratch.
As we build towards RSI, models must use ever-larger budgets efficiently.
To achieve this, we’re working on tasks that scale spend to $100,000 and beyond, allowing agents to perform the complete operations of a human-run lab.
We’re partnering with frontier labs to turn models into curious, capable researchers. Reach out to us if you want to add your model to our leaderboard.
Deep dive: frontierml.org/blog/
Human researchers decide everything from compute and data spend to what experiments to run. To replace them, models need this judgement.
We’re introducing FrontierML, a benchmark testing how well agents use money to scale model performance on 21 challenging, end-to-end ML tasks.
We’re introducing intrinsic discovery, a new learning method that allows models to explore and learn from open-ended environments without human supervision.
Using intrinsic discovery, we trained a world model that surpasses GPT-5.6 Sol at simulating terminal environments.
Induction Labs retweeted
always cool to see past @cohere teammates build great technology
many such cases
Induction Labs retweeted
i've been working with these guys building this in stealth for some time now. In fact they were one of the motivating reasons for YC to go and acquire its first GPU cluster.
I'm so excited that the whole world can get a glimpse at what they're building.
Induction Labs retweeted
Induction’s imagination models are a bet that internet-scale video can teach models how computers actually behave. It's a hugely ambitious vision and @jonat_li is exactly the kind of brilliant 19 year old who can pull it off.
Induction Labs retweeted
Despite only ever seeing computer video, Photon-1 learns general world concepts.
With a small amount of finetuning, it can predict billiard physics and checkers states better than baselines trained on the same data.
We’re introducing imagination models: a new foundation model architecture that unlocks learning from internet-scale video.
Our first imagination model, Photon-1, learned to use a computer by watching 18 years of screen recording video without action labels.
Enabling the approach is our vision encoder.
It compresses each frame into 960 discrete latent tokens occupying just 2.2KB—roughly 100× smaller than vision LLM representations—while preserving key detail. This makes autoregressive prediction of future states practical at scale.
In principle, imagination models can learn from any unlabeled video, like people doing work in the real world. We’re working on scaling our approach to these new domains.
More technical details on how it works: inductionlabs.com/news/scali…
