Pinned Tweet
On popular benchmarks, our 30B model matches systems 20-30x its size (gpt-5.4-xhigh, DeepSeek-V3.2, Kimi-K2.5), while using up to 95% fewer reasoning tokens than comparable 30/32B agentic LLMs.
The trick: don't just reason less, reason about the right things. A learned configurator decides when to simulate, how far ahead, and when to skip planning entirely.
Efficient reasoning is an allocation problem, not a compression problem.
Model and code are openly available.
Frontier LLMs are converging on efficient, adaptive reasoning. Opus 4.7 lets the model decide how deeply to reason. GPT-5.5 achieves strong results with fewer reasoning tokens.
We study a related but more structural question: what 𝗸𝗶𝗻𝗱 𝗼𝗳 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 should we adapt?
Last year in SiRA (upper figure), we showed that simulative reasoning (System II), which uses a 𝘄𝗼𝗿𝗹𝗱 𝗺𝗼𝗱𝗲𝗹 to evaluate consequences of actions, yields up to 124% improvement over reactive baselines (System I), and that strong reasoning models (o1, o3-mini) fail as planners without this structure.
In our new paper SR²AM (lower figure), we add a learned 𝗰𝗼𝗻𝗳𝗶𝗴𝘂𝗿𝗮𝘁𝗼𝗿 (System III) that self-regulates when to simulate, how far ahead, and when to skip planning entirely.
Efficient reasoning is not just shorter reasoning: it is better allocation of simulation.
Jinyu Hou retweeted
Everyone's been showing Fable and GPT-6 Astra driving robot arms.
We trained an open 27B Qwen model to do it, purely from its own attempts in simulation, with zero demonstrations.
And it can run locally at 220+ tok/s on an RTX 5090.
On a real robot it had never touched, it now succeeds in 18 of 30 tests, up from 0 before training. It also copies a painting and designs its own chemistry experiment.
It generalizes and follows instructions better than π0.5, and on LIBERO it comes close to Astra max.
All without losing its general capabilities, like math and coding.
Built on Qwen3.8-27B. We've open-sourced the model weights, code, and full training infra! Check out our blog at mit-han-lab.github.io/pear-b….
Jinyu Hou retweeted
I do a little surfing in my free time. The hardest call is *when* to pop up — too early and the wave rolls over you, too late and it’s already gone
So I asked PAN 2, the world model we trained at @IFM_AI, to show me the same wave, three different timings. It shows what wipes me out, and what leads to a clean ride
Simulate first, wipe out less #PAN2 #IFM #WorldModel #Simulation
Jinyu Hou retweeted
Introducing PAN 2: an open foundation world model for physical intelligence.
A language model predicts the next word. PAN 2 predicts the next world: what happens after each action, grounded in time, physics and space.
Towards AI that simulates “what if?” before it acts.
More details: ifm.ai/pan/
Jinyu Hou retweeted
Embodied First Exam at #CoRL Late Breaking Results
Tomorrow 12:00–1:30 PM
Griffin, Level 2, JW Marriott
Welcome to join us!
embodied-first-exam.ai
🤖📜 We will be presenting Embodied First Exam (embodied-first-exam.ai/), a comprehensive evaluation framework for coding agents on robotics tasks, at CoRL Late-Breaking Results at Griffin, Level 2, JW Marriott 12:00–1:30 PM.
Jinyu Hou retweeted
Contextualized gene representations work surprisingly well for predicting multigene perturbation effects.
I’m excited to introduce EMBER, a pretrained gene-centric foundation model designed specifically to learn context-specific gene representations. A joint work with @jmuiuc and @ericxing.
With innovations in single-cell pretraining and dataset curation, EMBER gene representations can be combined with a lightweight probe for single gene knockouts, genetic interactions (double knockouts), and even drug perturbations.
Both a 🧵 and a 🎥, got in front of a camera to share some highlights! (1/n)
Jinyu Hou retweeted
What should a robot remember? Can it learn to remember better than we tell it to?
Excited to share MEMI: Reverse-Engineering What Robots Should Remember
📈 97.0% end-to-end, 99.4% w/ low-level controller on RoboMME, nearly solving it for the first time
🧵👇
While the "Astra-as-policy" numbers everywhere right now, everyone is using their own harness, control interface and protocol. It is exciting to see the Embodied First Exam come into place as a shared bar.
Personal fave: the EmbodiedOlympics suite. Watching robotic agents attempt robot sports never gets old.
Introducing Embodied First Exam:
One line of code to run coding agents in 1,090 embodied tasks across 31 simulation suites and 44 robots.
Can your agent go far beyond pick-and-place:
Puzzles with spatial constraints;
Olympic sports with multi-robot collab;
Real-time eval, where the world won't pause;
Dynamic, agile, whole-body control;
AutoResearch with sim2sim
Try your own agent:
harbor run -d embodied-first-exam -a <your_own_harness> -m <your_own_model>
9
401
Jinyu Hou retweeted
Joining @openai next month!
after seeing people's reaction to Alisa's post about her experience, I also wrote down some of the surprising things I wish I know before my research scientist job search: yongzx.github.io/blog/2026/0…
We discussed foundational issues underlying how to make truly “agentive” systems.
arxiv.org/abs/2606.23991
// Critique of the Agent Model //
Finally, a paper that tries to define what an agent is and what agency consists of.
Good read overall. (great bookmark)
The word agent now covers everything from a for-loop with tool calls to speculative machine superintelligence.
Eric Xing and colleagues ask where automation ends, and agency begins.
Drawing on Descartes and on science-fiction portrayals of autonomous beings, they analyze agent architectures along five dimensions: goal, identity, decision-making, self-regulation, and learning.
The argument is that genuine agency requires these structures to hold together in a specific way. Great paper overall, providing a vocabulary for arguing about what is and is not an agent.
Paper: arxiv.org/abs/2606.23991
Learn to build effective AI agents in our academy: academy.dair.ai/
Our “Critique of Agent Model” paper has been now released on Arxiv!
We discussed foundational issues underlying how to make truly “agentive” systems.
arxiv.org/abs/2606.23991
2
175
Decoding the “Critique of Agent Model” Paper evoailabs.medium.com/decodin…
A companion paper to the critique of agent model paper.
arxiv.org/abs/2507.05169
Excited to share our new paper, "Critique of Agent Model," with Prof. @ericxing and @mdeng34. We analyze agent architectures along five dimensions of goal, identity, decision-making, self-regulation, and learning to examine what it takes for a system to approach genuine agency. A key finding is the distinction between agentic systems, whose competence resides in engineered workflows, and agentive systems, whose capabilities arise endogenously. Building on this, we propose the GIC architecture, combining hierarchical goal decomposition, identity evolution, simulative reasoning grounded in a separately trained world model, learned self-regulation, and self-directed learning from both real and simulated experience, for building agents that remain auditable, controllable, and under human oversight.
With the rise of LLM systems marketed as "coding agents", "AI co-scientists", etc. that promise to drive up productivity, and at the same time outcry of "existential" concerns that AI escaping human control with destructive power under a speculative "machine agency" against humans, there has been lots of confusion about “What is an agent?” and “What constitutes agency?” It has become essential to clarify where automation ends and agency begins.
Also recently, developments in world models, action models are trending to mixing future prediction/simulation and action/plan generation altogether within a single architecture such as a VLM, conflating reward-driven action selection with fidelity-driven next-state prediction, undermining the reliability of both planning and simulation.
In this paper we analyze agent architectures along the axis of goal, identity, decision-making, self-regulation, and learning, and argue that genuine agency requires these structures to be internalized within the system itself rather than assembled through external scaffolding. We propose a “Goal-Identity-Configurator” (GIC) architecture for a general-purpose agent model, combining hierarchical goal decomposition, identity evolution, simulative reasoning grounded in a separately trained world model, learned self-regulation, and self-directed learning from both real and simulated experience.
Auditability, controllability, and safety of systems that possess greater autonomy and "agency” but remain under human oversight, can be better built with the GIC architecture that offers transparency, modularity, and checkpoints.
@mdeng34 , @jinyuhou0
openreview.net/forum?id=6fDZ…
Jinyu Hou retweeted
Replying to @DrJimFan
This is really impressive! Agreed that significant parts of robot training decisions can and should be delegated to AutoResearch
What we see differently is in *what environment* the training occurs, and *how to decide* when to train, when to deploy, and when to retrain
We argue that fully autonomous agents should recursively self-improve in a *world-model-based simulator* to scalably capture the factors of variation; when to train, when to serve, and when to retrain should be the decision of a learned *configurator* that's part of the agent
We are currently working on these directions, more to come
More details in our recent paper “Critique of Agent Model” with Prof @ericxing and @jinyuhou0
Paper: openreview.net/forum?id=6fDZ…
Jinyu Hou retweeted
Excited to share our new paper "Critique of Agent Model" with Prof @ericxing and @jinyuhou0
Current LLM systems are often marketed as "coding agent" or other "agentic" tools, but what actually makes something an agent?
Our new paper draws a line: if the goals, identity, decision-making, behavior regulation, and learning live in external scaffolding rather than the system itself, it's automation, not agency
Recent World Action Models tend to mix future prediction and action generation in a single model, but where does the world end and where does the agent begin?
Our paper also argues that the World Model and the Agent Model are free to talk end-to-end, but should be trained separately. This avoid their conflicting objectives from undermining the reliability of planning and simulation
Based on the analysis, our paper proposes the Goal-Identity-Configurator (GIC) architecture for a general Agent Model *with* a World Model
GIC Agent Model combines hierarchical goal decomposition, identity evolution, simulative reasoning, learned self-regulation, and self-directed learning from both real and simulated experience, al using a separate World Model for prediction and simulation
Paper: openreview.net/pdf?id=6fDZYJ…
With the rise of LLM systems marketed as "coding agents", "AI co-scientists", etc. that promise to drive up productivity, and at the same time outcry of "existential" concerns that AI escaping human control with destructive power under a speculative "machine agency" against humans, there has been lots of confusion about “What is an agent?” and “What constitutes agency?” It has become essential to clarify where automation ends and agency begins.
Also recently, developments in world models, action models are trending to mixing future prediction/simulation and action/plan generation altogether within a single architecture such as a VLM, conflating reward-driven action selection with fidelity-driven next-state prediction, undermining the reliability of both planning and simulation.
In this paper we analyze agent architectures along the axis of goal, identity, decision-making, self-regulation, and learning, and argue that genuine agency requires these structures to be internalized within the system itself rather than assembled through external scaffolding. We propose a “Goal-Identity-Configurator” (GIC) architecture for a general-purpose agent model, combining hierarchical goal decomposition, identity evolution, simulative reasoning grounded in a separately trained world model, learned self-regulation, and self-directed learning from both real and simulated experience.
Auditability, controllability, and safety of systems that possess greater autonomy and "agency” but remain under human oversight, can be better built with the GIC architecture that offers transparency, modularity, and checkpoints.
@mdeng34 , @jinyuhou0
openreview.net/forum?id=6fDZ…
The outlook of an AI-driven Digital Organism (AIDO), such as a virtual cell (VC), has recently captivated much excitement and imagination from both AI and Biology communities, but there remain many open questions, in particular, what model presents the best path to realize an #AIDO or a #VC?
In this paper we present a definition of the virtual cell based on World Model — an architecture recently emerged in AI that supports advanced capabilities such as action-conditioned simulation, dynamic state-evolution, counterfactual reasoning, and long-horizon planning in complex dynamic environments. When applied to biological scenarios, a world model of the virtual cell is a generative model that simulates biological possibilities of a cell under any natural or artificial interventions and environments.
Such a virtual cell world model (VCWM) contrasts predictive foundation models on specific tasks, such as gene-expression perturbation prediction, as seen in some recent definitions of the virtual cell. At the same time not every biological foundation model built on sequence or structure or expression only can be repositioned as a world model if there is no multi- or pan-modality, stateful embedding, continuous action-conditioning, and dynamic state-transition and rollout.
We presents a novel architecture for #VCWM based on the GLP (generative latent prediction) framework that enables simulated cell as an end-to-end platform. Stay tuned for the release of the first implementation of VCWM from @genbioai soon.
@dasongle, @zivbj, @ElijahCole, @probablybots, @EuxhenH, @cfeinau, @deboramarks, @fabian_theis, @mmbronstein, @pkoo562
openreview.net/forum?id=hZNx…



















