Archive

14 digests

Tags

Jun 21, 202610 items

DeepSeek's $7.4B Series A redraws who bankrolls Chinese frontier AI — Tencent in, Alibaba and ByteDance out.

compute infrastructure policy vendor:deepseek china agent safety vendor:amazon

Jun 20, 20269 items

Anthropic's robodog rerun is the week's clearest capability-gain data point: Opus 4.7, fully autonomous, did the Project Fetch tasks ~20x faster than the fastest human team did a year ago.

vendor:anthropic agent robotics evaluation paper benchmark llm model-context-protocol

Jun 19, 20269 items

The US restricted Anthropic's new Fable model to US persons — and the real consequence is pushing the rest of the world toward Chinese open weights.

policy vendor:anthropic compute open-weight safety agent vendor:google evaluation

Jun 18, 202610 items

Z.ai's GLM-5.2 is now the best open-weights model on the planet, trailing only Claude Fable 5 on agentic web-dev.

open-weight llm moe benchmark vendor:zai agent infrastructure vendor:github

Jun 17, 202610 items

A 3B dense model lands AIME26 at 94.3 and claims parity with flagship reasoners, while OpenAI ships a method to preview model behavior before deployment.

paper llm evaluation open-weight reasoning safety vendor:openai agent

Jun 16, 20269 items

An OpenAI model autonomously solved an 80-year-old Erdős problem from a single prompt — the first time a prominent open question central to a math subfield was cracked by AI rather than merely assisted by it.

paper evaluation vendor:openai reasoning science policy safety vendor:anthropic

Jun 15, 202610 items

The White House forced Anthropic to suspend Claude 5 over a jailbreak-plus-cyber concern — the moment AI governance stopped being voluntary.

policy safety vendor:anthropic compute infrastructure agent evaluation paper

Jun 14, 20268 items

The US has ordered Anthropic to cut off all foreign nationals — including its own employees — from its frontier models on national-security grounds.

policy vendor:anthropic safety compute agent infrastructure open-weight llm

Jun 13, 20269 items

MiniMax's M3 stack arrives all at once — a sparse-attention kernel that cuts 1M-context attention compute 28x and a proof engine that clears human gold-medal thresholds on IMO and USAMO.

paper infrastructure llm open-weight compute evaluation reasoning agent

Jun 12, 20268 items

OpenAI buys Ona to give Codex a persistent place to run — the agent-infrastructure land grab is now louder than the model race.

agent infrastructure vendor:openai compute vendor:anthropic policy llm paper

Jun 11, 202610 items

Google ships DiffusionGemma, the first open-weight diffusion LLM fast enough to actually use — 1000+ tok/s on an H100, Apache 2.0.

open-weight llm vendor:google inference diffusion agent safety security

Jun 10, 20268 items

Anthropic shipped its smartest model twice — once with safety classifiers (Fable 5) and once without (Mythos 5) — and the silent degradation of 'frontier AI' requests is the real story.

llm vendor:anthropic safety alignment evaluation open-weight multimodal vendor:google

Jun 9, 20268 items

Apple's third-generation Foundation Models confirm it stopped trying to win frontier AI alone — the capable tier is Gemini-derived and runs on NVIDIA GPUs inside Google Cloud.

vendor:apple vendor:google llm on-device multimodal compute infrastructure vendor:nvidia

Jun 8, 20269 items

A genuinely quiet news cycle with no lab announcements, no model drops, and no compute deals — the only real signal is a cluster of papers shaving an order of magnitude off video and embodied-AI inference.

paper multimodal video efficiency llm robotics vla distillation