Introducing PC-ALM, a local-learning alternative to backpropagation. Our method trains 1000-layer neural nets using only local dynamics, and without backprop. Blog: Standard deep learning relies on backpropagation. The brain, however, cannot implement backpropagation, at least not exactly. How can a physical system, such as the brain, solve multilayer credit assignment without explicit use of backprop? We look for inspiration in two related fields: distributed optimization and NeuroAI. In NeuroAI, predictive coding asks each neuron activation to solve an energy-based inference problem instead of using a standard forward pass. That inference step can be implemented as energy-minimization dynamics on local prediction errors. This perspective -- each layer as a dynamical system -- has proven promising, but performance of predictive coding hasn't scaled well with depth. Credit signals at far ends of the network struggle to diffuse into internal layers. We turn to distributed optimization, generalizing predictive coding to use an augmented Lagrangian instead of energy. This motivation stems back to a classic 1988 paper by LeCun, showing that the Lagrange multipliers of a deep network can be identified with gradients of a supervised loss. The augmented Lagrangian then bridges LeCun's perspective to the standard predictive coding that is used in NeuroAI. We find that this new perspective yields a natural PC-like alternative to backpropagation, resulting in a method we call PC-ALM. PC-ALM differs from PC in that it introduces dual neurons (Lagrange multipliers) as part of the layer-local dynamics, resulting in each layer acting as a PI feedback control system to minimize local prediction errors. We find that PC-ALM is capable of propagating signals to seemingly arbitrary depth, especially in deep narrow networks where standard PC struggles to learn. Ultimately, our motivation here is to understand how distributed physical systems, such as the brain, can compute credit signals using only local coupling and local dynamics. PC-ALM may also inform deep learning in neuromorphic hardware, where dynamics are cheaper than on GPUs. Paper: Code:

Peak performance across hard benchmarks: • Best or joint-best on 5/8 benchmarks (DeepSWE, Chartography, Toolathon, GDP.pdf, SWEFish) • Chartography: 48.3 (outperforming Opus 5 & Fable 5) • DeepSWE: 74.3 Achieved without Fable 5, Fable 5.1, or GPT-6-Astra in the agent pool. Details: 🐡

Fugu Ultra v2 is now live on @OpenRouter 🐙 Our flagship orchestration engine built for peak performance on complex multi-step reasoning, autonomous research, and full-stack software development.

Introducing Fugu Max and Fugu Ultra v2: the next evolution of Sakana Fugu’s multi-agent orchestration system. Try: Blog: The frontier that actually matters is the Pareto frontier: capability on one axis, cost on the other. But the industry still treats it as a static menu of isolated models. Today we are resolving that with a dynamic architecture: Fugu Max expands the Pareto efficiency frontier. By orchestrating our largest pool of open-weights and specialized models to date, including NVIDIA Nemotron family, it dynamically routes tasks to the leanest capable model. Fugu Max delivers performance within striking distance of elite models at two to six times lower cost. Fugu Ultra v2 pushes the peak capability of orchestration higher than ever before. On Chartography, it outperforms Opus 5 and Fable 5. On DeepSWE, it outperforms models that cost three to five times more per token. Crucially, it does all of this without Fable 5, Fable 5.1, or GPT-6-Astra in its agent pool. The Fugu orchestration system evolved with model resiliency in mind. It does not rely on individual frontier models to deliver frontier output. By orchestrating a swappable pool of open and specialized models, it outperforms closed ecosystems while protecting users from vendor lock-in, API revocations, and sudden service cutoffs. Fugu Max expands the Pareto frontier outward. Fugu Ultra v2 pushes it upward. Orchestration does not force a choice between cost and capability. It advances both simultaneously.

After rigorous testing, our joint AI project with Daiwa Securities is entering the full-scale production phase. We're bringing our agentic AI systems to @Daiwa_JP’s wealth management teams to accelerate complex market analysis in volatile markets. Big milestone for Sakana AI!

We are excited to share our latest work, together with @nyuniversity: "Dream-Cubed: Controllable Generative Modeling in Minecraft by Training on Billions of Cubes." Blog: Paper: Code: Generative AI has made incredible progress in language modeling, far beyond other modalities, where words and tokens offer a natural compositional unit for scalable training. This is similar to Minecraft and many other popular video games, where developers rely on cubes, tiles, and other discrete primitives to build rich, interactive worlds. In this work, we show that using cubes as tokens allows large transformers to do the same. Our contribution is two-fold: 1/ We release Dream-Cubed to the research community, a large-scale dataset of Minecraft worlds designed for generative modeling. Our data comprises tens of billions of carefully-balanced cubes from procedurally generated Minecraft terrain and high-quality human-authored maps (obtained with the authors' consent). 2/ We use our data to train a family of powerful transformers for efficient generation of interactive 3D environments at cube resolution. We show how our models allow players to mold the world around them by generating structures, terrain, and maps that are immediately editable and playable. Using high-quality data, we demonstrate that these models can be successfully trained with different training objectives, including both continuous and discrete diffusion, unlocking targeted inpainting, large-scale outpainting, and user-conditioned generation of infinitely sized worlds with fine-grained block-level control.

Announcing the Claude Code-compatible interface for our new Fugu-Ultra v1.1! 🐡 Put a dynamically coordinated team of frontier models to work inside the coding workflow you already know. Instead of relying on a single model to write, debug, and execute your code, you can now orchestrate a diverse pool of state-of-the-art models directly from your terminal. Put the whole school to work on your next task: 🐟

Our team just shipped Fugu-Ultra v1.1! 🐡 By dynamically orchestrating the latest frontier models, we pushed performance up by 7.9 points. We are now beating Fable 5 in complex coding and reasoning tasks without even having Fable 5 in our agent pool. Collective intelligence is the future.

Announcing Fugu-Ultra v1.1 🐡 We’ve been thrilled by the reception to the Fugu model family. Thanks to everyone who tried it, shared feedback, and trusted Fugu with real work. Today, we’re releasing Fugu-Ultra v1.1 → Upgraded to incorporate the latest frontier models, resulting in stronger performance across every benchmark shown, including gains of up to 7.9 points over v1.0, with particularly strong results on ProgramBench and Terminal Bench 2.1. Fugu-Ultra v1.1 is more capable across coding, agentic tasks, and advanced reasoning, and available at the same price as Fugu-Ultra v1.0 The frontier keeps moving, and Fugu keeps getting better.

Can test-time scaling work for diffusion language models? In our #ICML2026# paper "UnMaskFork," we show that having multiple masked diffusion language models collaborate on a single answer improves performance on coding and math tasks. Blog: Test-time scaling is an actively researched technique that boosts LLM performance by using inference-time compute, for example, by having a model think longer or repeatedly refine its answers. This allows us to enhance performance simply by increasing computation during inference without relying on additional training, giving us the flexibility to balance compute costs and performance based on the specific use case. Unlike standard LLMs that generate text left-to-right, masked diffusion language models (MDLMs) generate text by gradually filling in a fully masked sequence. MDLMs can generate multiple parts of a sequence in parallel, offering potential speed-ups, and they can generate flexibly while seeing the entire sequence at once. This makes them an actively studied new paradigm in language modeling. We found that the standard LLM approach of "raising the temperature to increase randomness and generate diverse answers" does not work well for MDLMs like Dream-Coder. Instead of relying on this randomness, our proposed method, UnMaskFork (UMF), creates diversity through "model switching." Multiple MDLMs share the task of unmasking a single answer, and we use Monte Carlo Tree Search to search for a promising sequence in which different models handle different stages. Each model picks up where the others left off, filling in the parts it is most confident about. This collaborative approach allows us to explore diverse answers while maintaining generation quality, consistently outperforming existing test-time scaling methods on coding benchmarks and scaling effectively on math as well. Test-time scaling is also crucial for advancing MDLMs, and our work shows that UMF can sidestep the difficulties specific to them. UMF requires no additional training or changes to the models; it works simply by combining pre-trained models at inference time. This allows us to leverage the diversity of diffusion language models trained on different data and with different methods to improve performance. We believe the value of UMF will only grow as more diverse MDLMs emerge. This work is part of our broader research into "collective intelligence of AI," alongside methods like AB-MCTS and Sakana Fugu that have multiple LLMs collaborate. We'll continue pursuing research that turns model diversity into a source of strength. For details of the algorithm and illustrative examples showing how this collaboration works, please see our blog and paper. Paper: 🐟

Introducing Fugu-Cyber: an update to our Fugu orchestration model. It achieves state-of-the-art performance on real-world security benchmarks, matching cyber-focused frontier models like GPT-5.5-Cyber and Mythos Preview. 🐡

Sakana AI Teams With NVIDIA to Advance Open Model Innovation from Japan We're announcing the next phase of our collaboration with NVIDIA. We're bringing NVIDIA's open model stack, including the Nemotron family, into Sakana Fugu, our multi-agent orchestration system. Rather than relying solely on scaling individual monolithic models, our approach focuses on collective intelligence. Sakana Fugu operates as an intelligent orchestrator behind a single API, dynamically selecting, coordinating, and combining the strengths of multiple models for each task. This architecture keeps our system modular, adaptable, and resilient. As a natural next step to expand Fugu's capabilities, we're integrating NVIDIA Nemotron as a specialized agent, complementing the frontier and open models Fugu already orchestrates. Nemotron helps demonstrate how open models become far more useful when orchestrated within agentic systems rather than used in isolation. This collaboration creates a reinforcing cycle. Fugu gains a deeper pool of specialized capabilities, while NVIDIA can evaluate how its models perform when coordinated within complex, multi-step workflows. These real-world signals can continuously improve both the models and the orchestration layer. By combining Sakana AI's Japan-born collective-intelligence approach with NVIDIA's open models and accelerated computing, we aim to shape a future of AI that is modular, collaborative, and open by design.

We are pleased to share our latest research, now published in Nature Communications: “Smart Cellular Bricks: Physical Modules That Recognize Their Own Shape and Repair Themselves.” Blog: Paper: A long-running theme in our work is collective intelligence: the idea that sophisticated, robust behavior can emerge from many simple parts following local rules, with no central controller, as it does in a colony, a tissue, or a brain. We had mostly studied this in software and simulation. So this time we asked a simple question. Do the same decentralized principles hold up in the physical world, where communication is noisy and modules fail? To find out, we built a collection of simple cubic bricks. Each brick runs the same small neural network and talks only to the bricks it is physically connected to. No brick is told its position, or which shape it is part of. Yet from these purely local exchanges, the collective converges on the correct global shape, locates where modules are missing or damaged, and can even guide its own repair, inspired by how living tissue self-organizes and regenerates after injury. For us, this is a first step in a broader direction: taking the principles of collective intelligence we have studied in software and letting them emerge, decentralized and robust, in the physical world. In the future, we imagine smart materials that let structures sense and report damage on their own, and LEGO-like systems that recognize their own configuration and adapt in real time, pointing toward environments that are more robust, adaptive, and regenerative. This work is a collaboration between Sakana AI, IT University of Copenhagen and Autodesk.

Ren Ito, Co-Founder and Chairman of Sakana AI, has been appointed to the "AI for Good Global Commission", established by the United Nations and the ITU. As a Japanese AI company, we are honored to join global leaders, experts, and policymakers to actively contribute to building a trusted AI ecosystem, driving responsible innovation, and shaping the future of AI policy. Sakana AI 共同創業者の伊藤錬が、国連AI for Goodのグローバル委員会委員に就任しました。 日本のAI開発企業として本委員会へ参画し、信頼できるAIエコシステムの形成や責任あるイノベーションの支援に向けた国際的な議論に積極的に貢献していきます。

Fugu is now available on OpenCode! ✨ When our team was developing Fugu’s multi-agent orchestration, OpenCode was our tool of choice to verify our models. We share a core philosophy with the OpenCode team: the future of coding agents should be an open, collective ecosystem.

TechCrunch on Sakana Fugu and the broader conversation around frontier AI access in Asia. Our position remains that AI is best developed together rather than hoarded. Sakana AI will continue contributing to a more resilient ecosystem.