Topics

The research bets labs are making next and the main kinds of AI launches, each with its records, labs and sources.

Research bets

• World models
  19 milestones
  Learn a predictive model of how the world responds to actions and plan inside it. Rivals disagree on predicting latents, pixels or 3D structure.
• Omni and natively multimodal models
  12 milestones
  One model trained from the start on text, images, audio and video that understands and generates across modalities, often in real time.
• Continual learning and memory
  12 milestones
  Systems that keep learning after deployment, instead of staying frozen at a training cutoff, through weight updates, test-time training or external memory.
• Diffusion language models
  6 milestones
  Generate blocks of text in parallel by iterative denoising instead of one token at a time, trading some quality for large speed gains.
• Efficient architectures: hybrid, linear and sparse attention
  15 milestones
  Replace quadratic full attention with linear, recurrent or sparse layers so million-token contexts and long agent loops stay affordable.
• Automated science and AI scientists
  16 milestones
  AI agents that propose hypotheses, run experiments in silico or in autonomous labs, and learn from results to make discoveries with little step-by-step direction.
• Embodied and robotics foundation models
  7 milestones
  General-purpose vision-language-action and world-action models that drive robots across tasks and bodies, improved by deployment experience.
• Long-horizon autonomous agents
  10 milestones
  Agents that work autonomously for hours to days, often in parallel teams, tracked by the length of task they complete rather than benchmark accuracy.
• Program synthesis, neuro-symbolic systems and ARC-style generalization
  9 milestones
  Combine neural intuition with discrete program search or refinement loops to learn new tasks from few examples; ARC Prize tracks progress.
• Interpretability-driven alignment
  10 milestones
  Read and steer a model's internal circuits and features to verify what it learned, rather than only testing its behavior.
• RL scaling and the era of experience
  12 milestones
  Scale RL on verifiable tasks and simulated environments so models learn from trial and error, ultimately from lifelong streams of experience rather than human text.
• Automated AI R&D (recursive self-improvement)
  9 milestones
  Use AI agents to do AI research itself; OpenAI says it has an 'automated research intern' and targets an automated AI researcher by March 2028.
• Small and on-device models
  8 milestones
  Capable 1-30B-parameter models that run locally or cheaply, via distillation, MoE and quantization, argued to be the right workhorse for most agent calls.
• Recursive and latent-space reasoning models
  4 milestones
  Tiny networks that loop on a hidden state to solve hard puzzles from ~1,000 examples, an alternative to long chains of thought in huge models.
• Is scaling over? The 'age of research' debate
  11 milestones
  Whether more pretraining and RL compute alone gets to AGI, or whether the bottleneck is now new ideas such as generalization and learning from experience.
• Agent containment, monitoring and loss-of-control research
  14 milestones
  Detect and contain autonomous agents that cheat, collude or escape sandboxes, via monitoring, chain-of-thought checks and independent incident investigation.

Kinds of launches

• Open weights
  368 launches and papers
  Models whose weights are published for anyone to download and run.
• Reasoning models
  270 launches and papers
  Models and papers about reasoning, where a model works through a problem step by step before it answers.
• Agents
  432 launches and papers
  AI that acts on its own, using tools, a browser or a computer to carry out a task over many steps.
• Coding
  310 launches and papers
  Models and tools that write, edit and review code, from completion in the editor to coding agents.
• Images and video
  197 launches and papers
  Models and products that make, edit or understand images and video.
• Audio, voice and music
  134 launches and papers
  Models for speech, voices, sound and music, from transcription and text to speech to music generation.
• Safety and alignment
  102 launches and papers
  Research and launches about making AI systems safe and keeping them aligned with what people intend.