Skip to main content
Skip to categories
Crabhaus
What matters in AI, and why.
← Today's edition
Full wire feed
Fri 17 Jul · 14 stories · 5 sections
Frontier labs
3 items
Moonshot AI releases Kimi K3, a 2.8T-parameter model it says rivals Opus 4.8 and GPT-5.5; #1 on Frontend Code Arena, weights due July 27.
↗
A German AI consortium releases Soofi S, an open 30B model that tops several English and German benchmarks.
↗
NVIDIA releases Nemotron 3 Embed, an open embedding model that ranks #1 overall on the RTEB retrieval benchmark for agentic retrieval.
↗
Agents & tooling
4 items
OpenAI ships Codex Micro, a $230 Work Louder keypad whose keys accept, reject, and switch coding-agent tasks and tune reasoning levels.
↗
Google renames NotebookLM to Gemini Notebook and gives each notebook a secure cloud computer that can write and run code natively.
↗
LM Studio launches Bionic, an agent runtime that plans and executes multi-step tasks against locally hosted open models.
↗
1Password ships a macOS integration that lets Claude sign in to websites on a user's behalf without seeing the password or 2FA code.
↗
Research & papers
1 item
Analysis: swapping large models for small task-specific ones cuts inference cost only if many models share one GPU, not by size alone.
↗
Business & policy
4 items
The EU issues two DMA orders requiring Google to give rival AI assistants and search engines comparable access to Android and search data.
↗
A Chinese regulatory filing implies DeepSeek is valued at roughly $52B, after a fund deployed about $420M for an indirect 0.83% stake.
↗
Fireworks, which helps developers access AI chips and open models, raises $1.5B at a $17.5B valuation on over $1B annualized revenue.
↗
Filings show Anthropic CEO Dario Amodei gave $1M in May to Public First, a super PAC advocating AI-safety regulation.
↗
Also notable
2 items
NVIDIA unveils Cosmos 3 Edge, a world model that lets robots and AI agents perceive and navigate physical environments in real time.
↗
Meta's Oversight Board publishes its first evaluation of leading LLMs, warning that top AI models may be restricting free expression.
↗
Back to top ↑