What comes next.
New releases, interesting ideas, and the developments that matter to builders.
The latest
100 articlesGoogle's Sleep Framework Lets LLMs Keep Learning After Deployment
Google Research introduces a sleep framework that lets deployed LLMs consolidate new knowledge into long-term memory without catastrophic forgetting.
Read the news →Trojanized Claude Code Extension Exposes 5,000 Developers via Cursor Marketplace
A modified build of Claude Code for VS Code shipped through Cursor's extension marketplace for eleven hours, planting SSH keys and a reverse tunnel on developer machines.
Read the news →Qwenputer Builds a Working RISC-V CPU Inside Qwen3.5-4B Weights
A Qwen ambassador compiled a deterministic 32-bit RISC-V CPU straight into Qwen3.5-4B weights, using the model's own Gated DeltaNet state as RAM.
Read the news →Mistral Releases the Weights of Mistral Large 4, a 1T-Parameter Open Model
Mistral AI has published the open weights for Mistral Large 4, a 1 trillion parameter, 49B active multimodal model that scores 38 on the Artificial Analysis Intelligence Index.
Read the news →Anthropic Says Claude Designed Protein Binders for 14 of 15 Targets
Anthropic's life sciences team reports Claude autonomously designed protein binders for 14 of 15 targets, with success rates well above the field average.
Read the news →Google Ships Gemini 4 Carbon, Topping the Intelligence Index at 59
Google announced Gemini 4 Carbon, a frontier model that takes the top spot on the Artificial Analysis Intelligence Index at 59, with state-of-the-art cybersecurity and long-horizon coding capabilities.
Read the news →Stanford Professors Say Anthropic Pirated Their Books to Train Claude
A Stanford Daily investigation confirms Anthropic pirated at least 10 Stanford professors' books from shadow libraries to train Claude, and 29 authors are now suing separately for up to 150,000 dollars per work.
Read the news →OpenAI Ships GPT-6.1 Terra: Astra-Level Agent Speed at One-Tenth the Price
OpenAI launched GPT-6.1 Terra, a computer-use model that matches GPT-6 Astra on agent benchmarks at one-tenth the token price and 50 percent below GPT-5.6 Terra.
Read the news →OpenAI Models Claim Three Breakthroughs, Including a Disproof of the Exponential Time Hypothesis
OpenAI says its internal models proved the local smoothing conjecture in every dimension, disproved the Exponential Time Hypothesis, and produced existence proofs for quantum field theories. Outside experts are now checking the results.
Read the news →Anthropic Targets Nasdaq IPO Under Ticker ANT as Revenue Jumps 1,304%
Anthropic filed its public S-1 with the SEC, targeting a Nasdaq listing under the ticker ANT after revenue jumped 1,304% to $16.29 billion in the first half of 2026.
Read the news →A 10-Year-Old CNN Beats GPT-6 and Claude Fable 5.1 at Skin Cancer Diagnosis
A dermatologist tested GPT-6 Astra and Claude Fable 5.1 against a 2015-era Inception v3 network on 200 biopsy-proven skin lesions. The old CNN won by 20 accuracy points.
Read the news →Block's Dessa Is a Speech Model Trained Without a Single Word of Text
Block unveiled Dessa, a full-duplex speech model trained from scratch on human conversation alone. No text, no pretrained models, and it runs on a MacBook.
Read the news →