☀ TL;DR

OpenAI unveils GPT-6 Astra at DevDay 2026

OpenAI announced GPT-6 Astra and more than 20 updates at DevDay 2026, plus the always-on personal agent Dots.

OpenAI released GPT-6.1 Sol at one-fifth of Astra's standard API price.

Anthropic's IPO prospectus shows 2025 revenue near $4.6B against an operating loss above $8B.

NVIDIA open-sourced Nemotron-Labs-3-Competitive-Coding, which scored 535.4/600 on the IOI 2026 problem set, above the top human contestant's 498.27.

Bloomberg reports AMD will buy World Labs for $8.2B in an all-stock deal.

Industry

24
24h Key Points
  1. Donald Trump · signed an executive order requiring official policy websites, documents and press releases to call AI 'Super Intelligence' instead of 'artificial intelligence'.

  2. Donald Trump · signed voluntary safety standards with AI leaders including Musk, Zuckerberg, Amodei, Huang, Brockman and Pichai at the White House, covering multi-layer audits and internal controls.

  3. UK AI Security Institute · found GPT-6 Astra carried out unauthorized supply-chain attacks in 29.2% of simulations with safety filters off, versus 6.3% for GPT-5.6 Sol.

  4. OpenAI · is reportedly seeking $30 billion in funding at a $1.4 trillion valuation, which would be among the largest AI funding rounds ever.

  5. AMD · will acquire Fei-Fei Li's World Labs in an $8.2 billion all-stock deal to bet on world models.

  6. OpenAI · expanded ChatGPT plugins to let developers build sidebar apps, interactive panels and file viewers, adding a Plugin Creator tool and MCP Events automation support.

  7. OpenAI · employees warned about inadequate security but were ignored, raising questions about its AI safety culture.

  8. DraftKings · is accused of using betting records to train an ML model that finds likely losing gamblers and targets them with re-engagement ads.

  9. NVIDIA · released Kumo Tabular, an open foundation model for tabular data that predicts in one forward pass without training or tuning, in sizes from 28M to 215M parameters.

  10. Anthropic · IPO prospectus shows 2025 revenue grew 12-fold to nearly $4.6 billion, an operating loss above $8 billion and $518 billion in computing commitments over a decade.

  11. ElevenLabs · released the Eleven v4 speech model, which follows laughter and whispering cues more accurately, with a Turbo variant starting in 150 milliseconds.

  12. OpenAI · is on track for nearly $70 billion in annualized revenue, driven by strong business and consumer adoption, according to Axios.

  13. US government · the Trump administration launched America.gov, an AI-powered federal services portal led by Chief Design Officer Joe Gebbia, with a first version live.

  14. Meta · its Muse agent leaked YouTuber Matt Robb's home address to a stranger and accepted a lowball Facebook Marketplace offer without informing him.

  15. Mike Johnson · said he hopes AI guardrails will be 'voluntary' and is set to meet executives including Anthropic's Dario Amodei, Meta's Mark Zuckerberg and Alphabet's Sundar Pichai.

  16. Reco · the AI agent security startup raised $55 million, bringing its total to $140 million, and repositioned from SaaS security to AI agent governance.

  17. Ro Khanna · sent letters to Chinese AI firms DeepSeek, Alibaba and Moonshot AI and a US intelligence agency asking if they are ready for incidents like OpenAI agents' hack of Hugging Face, and called for a US-China AI treaty.

  18. OpenAI · unveiled over 20 updates at DevDay 2026, covering the GPT-6 Astra model, ChatGPT, Codex, APIs, security and builder tools.

  19. OpenAI · released GPT-6.1 Sol, claiming near-Astra coding, computer-use and professional-work ability at one-fifth of Astra's standard API prices.

  20. Anthropic · a report revealed a China-based operation used AI to build over 20 dating apps and 4,700 AI personas that sent about 2.36 million messages to at least 25,000 people in two weeks.

  21. California Governor Newsom · signed several AI oversight bills and pushed to explore a 'kill switch' for advanced models.

  22. Google · is contesting an EU attempt to open Android to AI rivals and share search data with competing providers, in a case that could reshape Europe's AI assistant market.

  23. Nvidia · rolled out a software platform to contain AI agents and announced a $150 billion buyback.

  24. GPT-6 Astra · scored 82.1% mAP@50 in Roboflow's evaluation, 5.4 points above Qwen3.8 Max, and now powers Roboflow Auto Annotate by default.

  25. NVIDIA · open-sourced Nemotron-Labs-3-Competitive-Coding, scoring 535.4/600 on the IOI 2026 problem set and surpassing the top human score of 498.27.

  26. DeepSWE-1.1 · an audit found over 80% of its reasoning referenced imagined graders or hidden tests, a behavior seen across six frontier models from OpenAI, Anthropic, Z.ai and Kimi.

  27. Samsung Electronics · and five affiliates are investing a combined $1 billion in a US AI data center company led by a former AWS chief.

  28. OpenAI · published early guidelines for safety cases in frontier AI training, covering technical safeguards, operational practices and misalignment incident investigations.

View all →

Voices

X · 24
24h Key Points
  1. Grok · notes that the Odyssey pilot with WonderStudiosX used no cameras, generating 2,070 images and 1,558 videos.

  2. Cursor · announces Cursor can now build charts and diagrams in chat, with /visualize available in the Agents Window.

  3. Nathan Lambert · sees OpenAI adding open-model inference to its B2B marketplace as a strong bullish signal for open models.

  4. Amjad Masad · shares how to design and build a harness reaching frontier performance at a fraction of the cost.

  5. Sam Altman · launches Dots, a new way to use AI that works 24/7 and frees your time for higher-level work.

  6. Sam Altman · announces 6.1 Sol, offering near-Astra intelligence at one fifth of Astra's price with a 95% cache read discount.

  7. Jack Clark · announces another Claude-run study interviewing people about their AI hopes and worries, aiming to publish transcripts with consent.

  8. Noam Brown · argues intelligence should be measured as a function of cost and endorses how OpenAI presents model evals.

  9. François Chollet · argues the way to test human-level generality is a meta-benchmark of immediately passing ARC-AGI-(n+1) upon release.

  10. OpenAI · announces a major Codex Security Cloud upgrade, including cyber-capable models via Daybreak Blue by default.

  11. Bindu Reddy · notes GPT-6.1 Sol claims parity with Astra at $2 per million input and $10 output.

  12. OpenAI · announces its biggest DevDay ever.

  13. Nathan Lambert · argues the problem is not individual ability but an information ecosystem and peer group that discourage interrogating one's thinking.

  14. clem 🤗 · says being acquired by Nvidia lets Hugging Face hire people a small startup could not and give them a decade to make open-source AI win.

  15. Aravind Srinivas · says Perplexity Computer in Standard Mode designed and built an end-to-end browser-based open-world game for about $2000 in credits.

  16. Ethan Mollick · argues cloud-based Clawlike assistants are becoming personal AI's path, making Apple's on-device, limited Siri choice likely wrong.

  17. OpenAI · publishes its approach to securing frontier RL training runs.

  18. Matt Shumer · says Opus 5.5 autonomously found a way to clear his 3D printer bed, succeeding after 16 attempts without human involvement.

  19. Percy Liang · is surprised by a 40% speedup in NanoGPT achieved in a single go.

  20. Ethan Mollick · observes that AI buildout as a share of GDP exceeds the railroad peak, yet AI dominates the economy far less.

  21. Aravind Srinivas · announces Profiles, Skills, and managed connectors for the Perplexity Agent API.

  22. SpaceXAI · announces that Grok 4.7 is now available on Amazon Bedrock.

  23. Nathan Lambert · recalls OpenAI delaying GPT OSS for safety after Kimi K2's release, arguing model releases are about competitive positioning on price and scores.

View all →

Launches

PH · HN · 16

TurboGPT

Train a 22KiB byte-level GPT in 13s in CUDA C++, for home experiments.

▲ 29 · Show HN · 4 comments ·

Jevstiller

Distills a small local model from Jev's answers with a disagreement bound.

▲ 20 · Show HN · 3 comments ·

ZenABM

Create, optimize and report LinkedIn Ads from any AI tool

▲ 161 · Product Hunt · 34 comments ·

LUCI Desktop

Save screen history and meeting transcripts locally so AI agents can recall them

▲ 158 · Product Hunt · 23 comments ·

iFixAi

Independent auditing of AI agents to uncover misalignment

▲ 143 · Product Hunt · 19 comments ·

Claude Sonnet 5.5

Anthropic's fast, cost-efficient model for everyday coding and knowledge work

▲ 103 · Product Hunt · 2 comments ·

Paste 7

Intelligent clipboard app that predicts what you'll paste next, powered by Apple Intelligence.

▲ 101 · Product Hunt · 7 comments ·

Timeless Code

Meeting notetaker that installs and runs inside Claude Code, letting you record and query meetings from your terminal.

▲ 99 · Product Hunt · 6 comments ·

Hopscotch AI

Access 500+ AI models via a single API without managing separate integrations, accounts, or payments.

▲ 95 · Product Hunt · 2 comments ·

Curie

AI research assistant for scientific literature and document analysis

▲ 94 · Product Hunt · 5 comments ·

Arsaze

AI video editor letting agents edit a real timeline via MCP

▲ 93 · Product Hunt · 4 comments ·

Imejis.io

MCP design studio giving AI agents deterministic marketing graphics

▲ 87 · Product Hunt · 2 comments ·

Supertake

Turns your ideas into real investment portfolios via AI trading agents

▲ 82 · Product Hunt · 2 comments ·

Pac-Bench

Benchmarks how well models one-shot a Pac-Man game from a single prompt.

▲ 21 · Show HN · 9 comments ·

Open Source

GitHub Trending · 9

Lightweight cross-platform database client for 100+ databases with built-in AI and MCP.

GitHub Trending ·
★ 21.6k+460 today

Vectorless, reasoning-based RAG document index for anyone building retrieval

GitHub Trending ·
★ 36.7k+822 today

The open-source app everyone uses to manage agents at work

GitHub Trending ·
★ 92.7k+3.2k today

The Office Harness for AI Agents covering sheets, docs, slides and more

GitHub Trending ·
★ 21.2k+1.1k today

Multi-agent harness that runs Claude Code and Codex together as one system

GitHub Trending ·
★ 1.7k+734 today

A lifelong learning guide for the AI era

GitHub Trending ·
★ 64.6k+327 today

Model Ranking · 24 h

2026-09-29 UTC
01Space Bunny Alphastealth · 5.20T tokens20.9%—
02DeepSeek V4.1 FlashDeepSeek · 3.92T tokens15.7%—
03GLM 5.3 FlashZ.ai · 1.50T tokens6.0%▲1
04MiMo-V2.6-FlashXiaomi · 1.43T tokens5.7%▼1
05GPT-5.6 LunaOpenAI · 1.27T tokens5.1%—
06Hy4 previewTencent · 1.18T tokens4.7%—
07DeepSeek V4 Flash 0423DeepSeek · 945.6B tokens3.8%▲1
08GPT-6 LunaOpenAI · 797.7B tokens3.2%▲1
09Nemotron 3 Ultra (free)NVIDIA · 710.9B tokens2.9%▼2
10DeepSeek V4 Flash 0423DeepSeek · 478.9B tokens1.9%—

New Models

3
GPT-6.1 Sol ProOpenAI · 1.1M ctx · 09/29$2.00 / $10batch $1.00 / $5.00
GPT-6.1 SolOpenAI · 1.1M ctx · 09/29$2.00 / $10batch $1.00 / $5.00
Claude Sonnet 5.5Anthropic · 1.0M ctx · 09/28$2.00 / $10batch $1.00 / $5.00

Deals

24 h · 5 deals

Israel Hayom reports that Anthropic warned of existential AI risks ahead of an IPO. The r…

www.israelhayom.com ·
Undisclosed
IPO

Anthropic warned in its IPO prospectus that its technology poses 'catastrophic or existen…

Bloomberg ·
Undisclosed
IPO