Open library
The essential reading of generative AI and agents, open in a clean reading view, with Meow and Standard summaries when available. No account needed to read.
560 papers · 558 with a Meow summary · 333 posts
Key publications from Anthropic, OpenAI, Google DeepMind, Meta, DeepSeek, Qwen, Z.ai, Kimi and other labs, plus the most influential essays and blogs.
David Silver, Demis Hassabis · Google DeepMind
AlphaGo Zero started from self-play instead of human game records, quickly surpassed human-level play, and beat the previously published champion-defeating AlphaGo 100 games to 0.
Julian Schrittwieser, Ioannis Antonoglou, Thomas Hubert, Karen Simonyan, Laurent Sifre, Simon Schmitt, Arthur Guez, Edward Lockhart, Demis Hassabis, Thore Graepel, Timothy Lillicrap, David Silver · Google DeepMind
MuZero learned to plan in Go, chess, shogi, and Atari without being given their rules, by modeling only what helps its decisions.
Google DeepMind
Gato uses one set of weights across games, image captioning, chat, and robot control, with context guiding what it outputs.
Daniel J. Mankowitz and Andrea Michi · Google DeepMind
AlphaDev searches CPU instructions for faster algorithms instead of merely refining familiar code—a promising approach when small routines are called again and again.
Alhussein Fawzi, Matej Balog, Bernardino Romera-Paredes, Demis Hassabis, Pushmeet Kohli · Google DeepMind
AlphaTensor turns algorithm discovery into a game: zeroing its tensor yields a provably correct matrix-multiplication algorithm.
Alhussein Fawzi and Bernardino Romera Paredes · Google DeepMind
FunSearch turns LLM proposals into evaluated code search: an automatic evaluator, not the model’s fluency, determines which ideas advance.
AlphaEvolve team · Google DeepMind
AlphaEvolve shifts LLM-assisted discovery from generating code to repeatedly testing and evolving programs; Google reports production computing uses and new mathematical results.
Google Research
ReAct interleaves a model’s reasoning with actions that query an environment, so fresh observations can guide what it does next.
Thinking Machines Lab · Thinking Machines
The proposal: build real-time collaboration into the model, so it can respond to a person’s changing input instead of waiting for neat turns.
Lilian Weng
An LLM agent’s autonomy comes from the system around the model: planning, memory, and tools. Those parts extend what it can do, but do not remove its core failure modes.
Simon Willison
For LLM engineering, an agent runs tools in a bounded loop to achieve a goal. It’s useful technical shorthand, not a universal meaning of “agent.”
Chip Huyen
A chain of almost-correct steps can still fail: the article says 95% accuracy per step falls to 60% over 10 steps and 0.6% over 100.
Shunyu Yao
The author calls the gap between AI’s benchmark achievements and its limited change to the world the “utility problem”—and argues that evaluation must address it.
LangChain Accounts · LangChain
An agent is not a yes-or-no label: the article ties autonomy to how much application control the LLM owns—and to the engineering support that level demands.
The Cognition Team · Cognition
The report argues that coding agents should be evaluated in full repositories, where they must locate files and use test feedback—not only solve isolated coding problems.
Anthropic
A rumor about a humorous passphrase sent investigators in the wrong direction; the recovered pattern pointed to timestamp-seeded lrand48 instead.
Anthropic
Opus 5.5’s sharpest pitch is cost per completed task: Anthropic reports competitive coding results against GPT-6 Astra and GPT-5.6 Sol at a fraction of the task cost.
Anthropic
Claude formalized Fermat’s Last Theorem in Lean in 11 days; the advance is computer-checking a known proof route, not discovering new mathematics.
Anthropic
Anthropic launches one underlying model with two safeguard profiles: Fable 5.1 is broadly available, while Mythos 5.1 is restricted. The company pairs performance and research claims with lower cache-read prices and tighter access controls.
Anthropic
Opus 5 raises reported coding and knowledge-work performance at Opus 4.8’s price, while its cybersecurity safeguards still block several operations.
Anthropic
Claude Opus 4.7 was about 20 times faster than the fastest human team on tasks participants had completed—but it still failed to fetch the ball reliably on its own.
Anthropic
The central deployment choice is not just model strength: Anthropic says Fable 5 and Mythos 5 share an underlying model, but their safeguards govern which requests each can answer.
Anthropic
Opus 4.8 pairs a model update with effort controls and dynamic workflows: users can tune response depth, while Claude Code can tackle larger jobs with parallel agents.
Anthropic
A more independent coding upgrade comes with a catch: old prompts may now be followed more literally, and token use can change.
Anthropic
Anthropic’s case is that Sonnet 4.6 brings broader coding, computer-use, and long-context capability to the Sonnet tier; Opus 4.6 remains its choice for the deepest reasoning tasks.
Anthropic
The important shift is toward longer, more complex work, with context and effort controls to help manage it.
Koray Kavukcuoglu · Google DeepMind
Google pitches Argon for long, multi-step professional work, but its first rollout is restricted to trusted cyber defenders while safeguards are strengthened.
Mateo Quiros · Google DeepMind
Computer use moves into Gemini 3.5 Flash, letting agents work across browser, mobile, and desktop; Google calls it its best performance yet, but the captured benchmark graphic provides no readable results.
Equipe do Google Gemini · Google DeepMind
Google presents Gemini 3.1 Pro as a core-reasoning upgrade for difficult tasks, citing a new-logic benchmark result; it remains in preview while Google validates updates.
Simon Willison
A one-line debugging request became a chain of browser tests and self-built instrumentation—Fable found ways to observe the bug without being explicitly told how.
Simon Willison
The “Claw” idea shifts attention from the model to the layer that makes an agent operate over time.
Ethan Mollick
The author’s main reversal: models may learn to organize agents without elaborate human-designed management.
Ethan Mollick
The author’s central claim: Claude 5 Fable makes ambitious work possible with little input, but shifts the user from directing the process toward commissioning its outcome.
Ethan Mollick
Choosing AI now means judging the model, the app, and the harness together: tool-equipped agents can take actions and complete multi-step work, not just chat.
Mistral
Mistral presents ML4 as an open-weight, natively multimodal model for coding, agents and security. Its benchmark results are from a preview; weights and fuller technical details are still forthcoming.
Latent.Space · Latent Space
Astra’s reported leap is from writing code to operating AI-engineering workflows; its ability to parallelize also makes the headline hourly cost conditional.
Z.ai
GLM-5.1’s central claim is that an agent can keep improving over long tasks by revising its strategy, not merely by starting stronger.
Z.ai
Z.ai presents GLM-5 as an open-source model for complex engineering and long-running agents, claiming best-in-class open-source performance in reasoning, coding, and agentic tasks.
Meta Superintelligence Labs · Meta
Muse Glimmer’s local-agent pitch depends on fitting a 30B model on consumer hardware and keeping generation responsive—not on leading every benchmark.
Peter Steinberger
The author is joining OpenAI to bring agents to everyone, while moving OpenClaw to a foundation intended to keep it open and independent.