Nvidia Alpamayo 2 Super, the Frontier Open Model for Autonomous Vehicles

6 hours ago

Autonomous vehicles struggle most with rare, complex 'long-tail' scenarios requiring situational understanding and real-time reasoning beyond object detection.
NVIDIA Alpamayo 2 Super is an open reasoning model for autonomous driving, built on Cosmos 3 Super Reasoner using reinforcement learning.
It is licensed under OpenMDW-1.1, a permissive open license that allows commercial use, fine-tuning, and redistribution.
Alpamayo 2 Super ranks first on the LingoQA autonomous driving reasoning benchmark, outperforming models like GPT-4o and Gemini 2.5 Pro.
The model delivers five key outputs: trajectory, chain-of-causation trace, meta-action, reasoning auto-labels, and visual QA with 2D grounding.
It supports a cloud-to-car workflow: frontier-scale reasoning in the cloud for development and efficient inference in production vehicles.
Alpamayo 2 Super is part of a broader open ecosystem including simulators (AlpaSim), RL frameworks (AlpaGym), and open datasets.

Inkling-Small is a general-purpose multimodal model that accepts text, image, and audio inputs and generates text outputs.
It can be used with libraries like Transformers, vLLM, SGLang, and Docker, as well as through APIs and third-party providers.
The model uses a 42-layer decoder-only transformer with a sparse Mixture-of-Experts (MoE) architecture, totaling 276B parameters but activating only 12B per token.

Laguna S 2.1 is a 118B-parameter Mixture-of-Experts model with 8B active parameters per token and a 1M-token context window, developed in under nine weeks.
It achieves competitive scores against much larger models on long-horizon coding benchmarks, scoring 70.2% on Terminal-Bench 2.1 and 40.4% on DeepSWE v1.1.
All evaluation trajectories are publicly released at trajectories.poolside.ai for transparency and reproducibility.

Nanbeige4.2-3B is a compact agentic model with 3B non-embedding parameters, using a Looped Transformer architecture to increase capacity without adding parameters.
It outperforms larger models like Qwen3.5-9B and Gemma4-12B on general-agent, code-agent, and reasoning benchmarks.
The model is suitable for local deployment as a personal assistant, excelling in daily, office, and deep research tasks.

Transformer Transformer is a unified model that generates a complete robot embodiment (links, joints, motors, inertia) optimized for a given manipulation demonstration.
The model uses RoboTokens, a unified tokenization of robot embodiments, states, and actions, enabling it to span diverse embodiment spaces like wheeled bimanual, quadrupeds, and humanoids.
It is a diffusion transformer trained with DDIM, and by masking different tokens it can act as a generator, cross-embodiment controller, or motion-conditioned designer.

Subtext reveals the internal predictions of a language model at multiple layers, showing what words the model is leaning toward before it outputs them.
It uses a Jacobian lens to decode hidden states, allowing users to observe the model's decision-making process in real time.
Key observations include the model settling on a conclusion before it speaks, planning ahead, and displaying multi-step reasoning like a 'two-hop' chain.

Inkling-Small is a general-purpose multimodal model that accepts text, image, and audio inputs and generates text outputs.
It can be used with libraries like Transformers, vLLM, SGLang, and Docker, as well as through APIs and third-party providers.
The model uses a 42-layer decoder-only transformer with a sparse Mixture-of-Experts (MoE) architecture, totaling 276B parameters but activating only 12B per token.

Laguna S 2.1 is a 118B-parameter Mixture-of-Experts model with 8B active parameters per token and a 1M-token context window, developed in under nine weeks.
It achieves competitive scores against much larger models on long-horizon coding benchmarks, scoring 70.2% on Terminal-Bench 2.1 and 40.4% on DeepSWE v1.1.
All evaluation trajectories are publicly released at trajectories.poolside.ai for transparency and reproducibility.

Nanbeige4.2-3B is a compact agentic model with 3B non-embedding parameters, using a Looped Transformer architecture to increase capacity without adding parameters.
It outperforms larger models like Qwen3.5-9B and Gemma4-12B on general-agent, code-agent, and reasoning benchmarks.
The model is suitable for local deployment as a personal assistant, excelling in daily, office, and deep research tasks.

Transformer Transformer is a unified model that generates a complete robot embodiment (links, joints, motors, inertia) optimized for a given manipulation demonstration.
The model uses RoboTokens, a unified tokenization of robot embodiments, states, and actions, enabling it to span diverse embodiment spaces like wheeled bimanual, quadrupeds, and humanoids.
It is a diffusion transformer trained with DDIM, and by masking different tokens it can act as a generator, cross-embodiment controller, or motion-conditioned designer.

Subtext reveals the internal predictions of a language model at multiple layers, showing what words the model is leaning toward before it outputs them.
It uses a Jacobian lens to decode hidden states, allowing users to observe the model's decision-making process in real time.
Key observations include the model settling on a conclusion before it speaks, planning ahead, and displaying multi-step reasoning like a 'two-hop' chain.

Hasty Briefsbeta