Search

CAR-bench is a new 'driving test' for AI assistants that checks if they can stay careful, honest, and consistent during real back-and-forth conversations in a car.

The paper teaches large language models to learn from detailed feedback (like error messages) instead of only a simple pass/fail score.

This paper explains how to turn large language models (LLMs) from quiet students that only answer questions into active agents that can plan, act, and learn over time.

LongCat-Flash-Thinking-2601 is a huge 560-billion-parameter Mixture-of-Experts model built to act like a careful helper that can use tools, browse, code, and solve multi-step tasks.

This paper fixes a common problem in reasoning AIs called Lazy Reasoning, where the model rambles instead of making a good plan.

Step 3.5 Flash is a huge but efficient AI that keeps 196 billion total parameters but only wakes up about 11 billion per token, so it thinks smart and fast.

Nemotron 3 is a new family of open AI models (Nano, Super, Ultra) built to think better while running faster and cheaper.

INTELLECT-3 is a 106B-parameter Mixture-of-Experts model (about 12B active per token) trained with large-scale reinforcement learning and it beats many bigger models on math, coding, science, and reasoning tests.

AdaReasoner teaches AI to pick the right visual tools, use them in the right order, and stop using them when they aren’t helping.

Agentic AIs don’t just chat; they plan, use tools, and take many steps, so one wrong click can cause real harm.

Small AI models often stumble when a tool call fails and then get stuck repeating bad calls instead of fixing the mistake.

AgentVista is a new test (benchmark) that checks whether AI agents can solve tough, real-life picture-based problems by using multiple tools over many steps.

The paper teaches AI agents better by grading not just their final answers, but also how they think and use tools along the way.

MobilityBench is a big, carefully built test that checks how well AI helpers can plan real-world routes using natural language and map tools.

The paper builds a new way to create realistic, long conversations between people and AI that use tools like databases.

KARL is a smart search helper that learns to look up information step by step and explain answers using the facts it finds.

Monadic Context Engineering (MCE) is a way to build AI agents using math-inspired Lego blocks called Functors, Applicatives, and Monads so state, errors, and side effects are handled automatically.

This paper builds a new test, LongShOTBench, to check if AI can truly understand long videos by using sight, speech, and sounds together.

PyVision-RL teaches vision-language models to act like curious agents that think in multiple steps and use Python tools to inspect images and videos.

WideSeek-R1 teaches a small 4B-parameter language model to act like a well-run team: one leader plans, many helpers work in parallel, and everyone learns together with reinforcement learning.