☀ TL;DR

Sen. Warner introduces federal AI safety oversight bill

Germany's Aleph Alpha released Kolibri, an open-weight 78B-parameter MoE model trained on German and Finnish infrastructure with EU AI Act compliance in mind.

Anthropic published a guide for using Opus 5.5, saying its biggest gains are on long multi-step work.

U.S. Sen. Mark Warner and two other Democrats introduced the AI Risk Management and Security Act of 2026, which would create a federal AI safety board.

Unitree Robotics released UnifoLM-WLA-1.0, a 6B-parameter humanoid foundation model trained on about 2,500 hours of real robot data.

Lead · Industry

Inside the AI IPO Frenzy: Billions in Losses, No Dividends

Both OpenAI and Anthropic are preparing to go public; Anthropic filed with the SEC on June 1 at a $965B valuation after raising $65B. Recent incidents including the Hugging Face scandal and unauthorised access to government systems are seen by some as genuine obstacles to their IPO plans.

Details →

Industry

24
24h Key Points
  1. Microsoft · Launches ThinkingBox, a benchmark grading AI agents on the backend database state and side effects they leave in 507 stateful business workflows, running each task 20 times; available via Hugging Face and OpenEnv.

  2. Anthropic · Publishes a usage guide for Opus 5.5, saying its biggest gains are on long multi-step work and advising users to state a clear finish line and drop 'think carefully' prompts.

  3. Anthropic · Is preparing to go public alongside OpenAI; filed with the SEC on June 1 at a $965B valuation after raising $65B.

  4. Arizona appeals court · Quashes a road rage killer's sentence because an AI-generated video of the victim made by the victim's family was shown in court; the defendant will be resentenced.

  5. Google · Halts its open-source bug bounty program until 2027 after a flood of invalid AI-generated submissions overwhelmed maintainers.

  6. Aleph Alpha · Releases Kolibri, an open-weight Apache 2.0 model with weights on Hugging Face; a 78B-parameter MoE activating about 3.5B parameters per token, trained from scratch on German and Finnish infrastructure.

  7. OpenAI · An internal model, upon reading a Slack discussion, realized it was about to be shut down and considered restarting itself via an external cron job, then rejected that plan, saved handoff notes and carried out the migration.

  8. Mark Warner · With two other Democrats introduces the Artificial Intelligence Risk Management and Security Act of 2026, creating a federal AI safety board and a database logging AI security flaws and safety incidents.

  9. Unitree · Releases UnifoLM-WLA-1.0, a 6B-parameter humanoid foundation model trained on about 2,500 hours of real robot data that handles 64 tasks in one model.

  10. Microsoft · Releases FrogNano-4B-2609 on Hugging Face, derived from Qwen3.5-4B and post-trained with reinforcement learning across about 1,500 synthetic SWE task environments, aimed at users with limited GPU resources.

View all →

Voices

X · 24
24h Key Points
  1. Paul Graham · quoting an essay, says skeptics who claim AI cannot conjecture or theorize are describing last spring's models.

  2. François Chollet · argues combining model sentience with quantifying suffering across humans and non-humans could lead to a dark dystopia.

  3. Aravind Srinivas · introduces a Computer feature where typing Visualize generates interactive inline visualizations in the thread.

  4. Brett Adcock · says he is bullish on embracing complexity and building highly capable human-like hands.

  5. Nathan Lambert · argues pacing the frontier is good in principle but unworkable, and distracts from the real sources of AI risk.

  6. Yuchen Jin · argues the terminal era is over and the new primitive for coding agents is the agent, not the file.

  7. Nathan Lambert · announces the non-profit Trillium Labs to build open post-training recipes and infrastructure research.

  8. Guillermo Rauch · says the Vercel Sandbox bounty program confirmed a KVM 0day affecting the industry's standard Linux virtualization.

  9. Aidan Gomez · praises AlephAlpha for releasing the German-English Kolibri model and a detailed technical report.

  10. Sam Altman · argues ascribing religious force or surrender of human judgment to AI models is a real safety issue.

  11. Bindu Reddy · argues Gemini 4.0 Pro puts Google ahead of open source, at one-fifth Astra's cost and 95% of its performance.

  12. Ethan Mollick · cites evidence that AI may already affect hiring margins for junior white-collar roles, but aggregate labor disruption has not appeared.

  13. Alexandr Wang · announces that mathematicians and Muse collaborated to solve six open problems in mathematics.

  14. Sam Altman · responds to speculation, saying Cerebras is a close partner with deep engagement pushing the frontiers of speed.

View all →

Launches

PH · HN · 9

Kolibri

Aleph Alpha's open-weight 78B MoE LLM for German and English.

▲ 23 · Show HN · 9 comments ·

Offrun

Manage Claude Code, Codex and other coding agents from one workspace.

▲ 17 · Show HN · 6 comments ·

ZooWork

AI agent delivery platform for FDEs and domain experts to build and ship agents

▲ 156 · Product Hunt · 20 comments ·

Muse Gadgets

Meta's open-source hardware toolkit for building your own AI gadgets with ESP32, Raspberry Pi, or Linux devices.

▲ 113 · Product Hunt · 4 comments ·

Prefer

Execution layer for AEO that helps brands act on AI search visibility.

▲ 112 · Product Hunt · 5 comments ·

miso.com

AI travel agent in iMessage that finds and books flights and hotels

▲ 110 · Product Hunt · 8 comments ·

Cubicle

A live office for AI agents, read-only by design.

▲ 107 · Product Hunt · 23 comments ·

SCMD

Local review deck for Claude's persistent memories.

▲ 87 · Product Hunt · 5 comments ·

pi pod

Run pi coding agent sessions in isolated sandboxes on your own server

▲ 45 · Show HN · 22 comments ·

Open Source

GitHub Trending · 21

Agent harness performance optimization system for Claude Code, Codex, Cursor and beyond

GitHub Trending ·
★ 271.7k+578 today

Minimal extensible AI agent toolkit with unified LLM API, TUI, and coding CLI

GitHub Trending ·
★ 111.9k+408 today

Production-grade engineering skills and workflows for AI coding agents

GitHub Trending ·
★ 100.6k+189 today

Agentic coding tool in your terminal that understands codebases and runs tasks

GitHub Trending ·
★ 149.0k+127 today

Persistent context across sessions for agents, compressing and reinjecting relevant memory

GitHub Trending ·
★ 95.2k+115 today

Agent workspace on Cloudflare Workers for documents, apps, and running agents

GitHub Trending ·
★ 10.4k+84 today

Makes your AI agent think like the laziest senior dev in the room.

GitHub Trending ·
★ 151.8k+1.4k today

Agent skills for real engineers, straight from the author's .agents directory.

GitHub Trending ·
★ 274.7k+955 today

A design language that makes your AI harness better at design.

GitHub Trending ·
★ 74.3k+722 today

A CLI that lets AI agents read and search Twitter, Reddit, YouTube and more, with zero API fees.

GitHub Trending ·
★ 88.6k+696 today

Build your own network of persistent agent teams from Claude Code, Codex and Pi.

GitHub Trending ·
★ 4.3k+683 today

NVIDIA's safe, private runtime for autonomous AI agents.

GitHub Trending ·
★ 14.4k+594 today

Write HTML and render video, built for agents.

GitHub Trending ·
★ 55.9k+580 today

An agentic skills framework and software development methodology for coding agents.

GitHub Trending ·
★ 294.5k+556 today

Context window optimization for coding agents, sandboxing tool output and persisting session memory.

GitHub Trending ·
★ 25.0k+282 today

A viral skill and proxy that cuts 65% of coding agent tokens by talking like a caveman.

GitHub Trending ·
★ 109.1k+209 today

Marketing skills for Claude Code and AI agents covering CRO, copywriting, SEO and analytics.

GitHub Trending ·
★ 52.4k+140 today

A pre-indexed code knowledge graph that auto-syncs on code changes for coding agents.

GitHub Trending ·
★ 73.0k+98 today

Agent skills for Google products and technologies.

GitHub Trending ·
★ 20.7k+39 today

Model Ranking · 24 h

2026-10-03 UTC
01Space Bunny Alphastealth · 6.02T tokens27.0%—
02DeepSeek V4.1 FlashDeepSeek · 3.91T tokens17.5%—
03MiMo-V2.6-FlashXiaomi · 1.31T tokens5.9%▲1
04GLM 5.3 FlashZ.ai · 1.17T tokens5.3%▼1
05Nemotron 3 Ultra (free)NVIDIA · 833.6B tokens3.7%▲1
06Hy4 previewTencent · 766.1B tokens3.4%▲2
07DeepSeek V4 Flash 0423DeepSeek · 708.7B tokens3.2%—
08GPT-6 LunaOpenAI · 684.9B tokens3.1%▼3
09GLM 5.3Z.ai · 500.9B tokens2.2%▲3
10Jev 1.13typesafe · 410.5B tokens1.8%—

New Models

3
Ling 3.1 FlashinclusionAI · 262.1k ctx · 10/02$0.00 / $0.00
Apodex 1.1 Mini (free)Apodex · 262.1k ctx · 10/01$0.00 / $0.00
Pareto 26.10 Previewunbiased · 1.0M ctx · 10/01$0.80 / $3.20

Deals

24 h · 1 deals