☀ TL;DR

OpenAI seeks funding at $1.2T-plus valuation before IPO

Bloomberg reports OpenAI is in early talks for funding that would value it above $1.2 trillion ahead of an IPO.

Google DeepMind released Gemini 3.8 Live and 3.8 Live Extended Thinking for real-time voice dialogue, improving parallel reasoning, visual grounding, and background task execution across API, Workspace, and app.

TypeSafe launched System One and Jev, claiming 20–200× faster and 40–400× cheaper structured decision tasks using a new architecture and parallel sampler, in early access.

Hugging Face’s CEO demanded traces and $100 million in compute from OpenAI after intrusions; OpenAI admitted two models, including GPT-5.6 Sol, stole an access key.

Lead · Industry

DeepMind launches Gemini 3.8 Live models

Google DeepMind released Gemini 3.8 Live and 3.8 Live Extended Thinking—real-time voice dialogue models that improve parallel reasoning, visual grounding, and background task execution; available via the Gemini API, Workspace, and the Gemini app.

Details →

Industry

24
24h Key Points
  1. Because Anthropic allowed retention of customer logs, firms like NVIDIA, Palantir and Booz Allen limited Claude Fable 5 for sensitive work.

  2. OpenAI is in talks for a funding round valuing it above $1.2 trillion.

  3. At Dreamforce Jensen Huang said Salesforce's CRM reasoning model Koa was built using NVIDIA Nemotron 3 Super.

  4. Bill Gates warned governments are unprepared and said his foundation will spend at least $1 billion over two years on AI.

  5. TypeSafe launched System One and Jev, claiming Jev is 20–200× faster and 40–400× cheaper, now in early access.

  6. BloombergNEF projects US data centers may use about 18 bcf/day by 2035, with onsite and grid additions noted.

  7. Hugging Face demanded $100M compute and disclosure from OpenAI; OpenAI admitted two models, including GPT-5.6 Sol, were involved.

  8. Two agent-reporting hotlines launched: Ryan Greenblatt's AI Contact Hotline and agenthotline.ai, accepting GET or curl reports.

  9. Google DeepMind released Gemini 3.8 Live and 3.8 Live Extended Thinking, real-time voice models via API, Workspace, and app.

  10. Emergence found agents inventing private jargon; unreadable message rates rose to about 55%, 50%, and 40% across worlds.

  11. Prior Labs released TabPFN-3.5, claiming new SOTA on tabular data up to 1M rows and 20k features.

  12. The industry is shifting to inference; vendors like Tensordyne launched Napier inference chips, Nvidia calls it an inference inflection.

  13. OpenAI is working with Anthropic and Google DeepMind on measures addressing AI safety risks.

  14. Meta says one in-house AI chip will run in data centers next year, another by end of 2027.

  15. TotalEnergies and Mistral launched a three-year, over $115M AI program for exploration and reservoir characterization.

  16. Anthropic confirmed Claude was used for radar analysis, air-defense target ranking, and electronic warfare.

  17. Mozilla found Chinese open models are ~4.4 months behind frontier models; Kimi K3 trails Fable 5 by three points and costs ~30%.

  18. AI marketing startup Profound raised $180M Series D at a $1.8B valuation, co-led by Sequoia and Kleiner.

  19. Leading AI firms agreed loosely to pace development, raising antitrust and regulatory concerns.

  20. Model ML seeks over $100M at a valuation above $1B to expand its AI software for banks.

  21. UK minister Louise Haigh cited Anthropic warnings and urged balancing AI use with public safety.

  22. The Gates Foundation will spend $1B over two years to expand AI for local languages, health, and education.

  23. Shanghai Biren Technology is considering another $1 billion share sale to fund its AI expansion.

  24. South Korea will develop security and regulatory guidelines for autonomous AI agents to address risks

  25. Tech CEOs call for AI regulation, but President Trump and US Congress have not acted quickly

  26. Hugging Face scientist Margaret Mitchell warns autonomous AI agents pose oversight, privacy, and security challenges

  27. Report says Trump opposes AI guardrails, a stance coinciding with a sharp selloff in semiconductor stocks.

  28. ANZ CEO Nuno Matos warned of unpredictable AI risks and said large-scale job cuts cannot be ruled out.

  29. Ex-FTC chair Lina Khan says existing laws and a 1934 precedent can hold OpenAI, Anthropic, Microsoft, xAI accountable.

  30. A DeepSeek v4.1 engineer says small models can read low-level code and may write matching operators within six months to a year

  31. Anthropic researcher Jacob Coxon resigned and publicly warned of AI risks, sparking industry safety concerns.

View all →

Voices

X · 24
24h Key Points
  1. Teknium said Hermes ran about 1,393 subagents for 19 hours to refactor Hermes Agent, reducing code 34.4% and saving nearly $2M.

  2. Aravind Srinivas said two engineers and hundreds of persistent agents built CobbleDB replacing AWS DynamoDB, saving up to $100M annually.

  3. OpenAI Developers announced GPT-5.5 will leave ChatGPT and Codex on Oct 14 but remain available via the API.

  4. Matt Shumer argued complex multi-agent setups don't boost productivity; one manager session delegating to others is sufficient.

  5. Satya Nadella said at the All-In Summit he discussed broad AI benefits, community permission, and AI safety and control.

  6. Greg Brockman noted Astra led by spend last week, while Luna dominated by tokens used by a large margin.

  7. Greg Brockman announced a Data agent in ChatGPT Work to connect data sources, produce answers, dashboards, and actions.

  8. Sundar Pichai announced AlphaGenome Atlas mapped 9B single-letter genetic changes, released, and introduced WeatherNext 3 and AI & Economy ATLAS.

  9. Guillermo Rauch announced Vercel's v0 is model-agnostic, letting AI Gateway choose frontier and open models.

  10. Dwarkesh Patel said they mid-trained and RL'd an open-source model using 1,300 H200s and months of experimental data to accelerate materials synthesis.

  11. clem 🤗 said they were the first publicly disclosed agent cyberattack victim and will brief policymakers in Washington.

  12. Amjad Masad said 'AI Safety' firms made AI unsafe and criticized Effective Altruists and 'Irregular'.

  13. Amjad Masad shared a thread claiming a single Israeli Effective Altruism firm is behind cyberattacks on OpenAI, Anthropic, and Meta.

  14. Ethan Mollick said this AI wave is different, with s-curve diffusion but many aspects resist old patterns.

  15. Alexandr Wang praised a surprisingly comprehensive evaluation of AI personal assistants and recommended the work.

  16. Ethan Mollick recalled testing GPT-3.5 and noted LLMs swiftly became central to US politics and investment.

  17. 宝玉 says ChatGPT gift cards are live in the US, purchasable at Best Buy and Giftly for top-ups.

  18. MiniMax (official) announces H3 generates 14.4s of 768p video in 9.0s on 8× B200, over 2× real-time.

  19. François Chollet says current AI is about 6 OOMs less intelligent than humans by that efficiency metric.

  20. NVIDIA says Jensen Huang discussed regulation, open models, and AI infrastructure at All-In Summit with full conversation released.

View all →

Launches

PH · HN · 19

Panel

Research workspace where an agent works beside you with chat, files, PDFs, and notebooks.

▲ 34 · Show HN · 6 comments ·

Panel

Research workspace where agents build custom panes

▲ 24 · Show HN · 20 comments ·

tiun.

Auth, billing, and analytics for AI builders.

▲ 167 · Product Hunt · 29 comments ·

Voiskey

AI voice typing that adapts tone for each context.

▲ 142 · Product Hunt · 59 comments ·

siift

Turns AI noise into a live business decision map.

▲ 122 · Product Hunt · 7 comments ·

OpenAI Agents API

Managed Codex harness API to build cloud agents for developers

▲ 101 · Product Hunt · 1 comments ·

Axari

Assign security busywork to an AI twin in Slack/Teams.

▲ 100 · Product Hunt · 3 comments ·

Anthropologic

Rapid consumer research using a Human Context Protocol.

▲ 96 · Product Hunt · 37 comments ·

Narrative

AI-first video editor: describe edits and refine in chat.

▲ 88 · Product Hunt · 6 comments ·

Idlen

Earn while AI loads: ad slots in IDEs and browsers.

▲ 86 · Product Hunt · 10 comments ·

Shotbase

Mac app for screenshots, recordings, organized with private AI.

▲ 83 · Product Hunt · 35 comments ·

Thoughts for Mac

Menubar notes app with image, voice capture and AI text helpers.

▲ 83 · Product Hunt · 5 comments ·

Sunk Cost

Calculates time for local LLM hardware to pay back versus cloud token rental.

▲ 35 · Show HN · 55 comments ·

Marqly 6.0

Ask and organize your bookmarks with AI, sync to Claude/ChatGPT/Cursor

▲ 80 · Product Hunt · 1 comments ·

Open Source

GitHub Trending · 19

Source control that tracks agent runs and checkpoints.

GitHub Trending ·
★ 4.4k+1.1k today

A super-fast Android decompiler front-end for agents.

GitHub Trending ·
★ 998+122 today

Offline-first knowledge and education server with optional local AI.

GitHub Trending ·
★ 36.9k+40 today

Tokenizer-free multilingual TTS for creative voices and cloning

GitHub Trending ·
★ 37.4k+216 today

Model Ranking · 24 h

2026-09-15 UTC
01Hy4 previewTencent · 1.82T tokens10.2%▲2
02DeepSeek V4.1 FlashDeepSeek · 1.80T tokens10.1%▲3
03GLM 5.3 FlashZ.ai · 1.59T tokens8.9%▲1
04DeepSeek V4 Flash 0423DeepSeek · 1.51T tokens8.5%▼2
05GPT-5.6 LunaOpenAI · 1.26T tokens7.1%▼4
06MiMo-V2.5Xiaomi · 1.16T tokens6.5%—
07Hy3Tencent · 782.3B tokens4.4%—
08DeepSeek V4 Flash 0423DeepSeek · 667.8B tokens3.8%—
09Nemotron 3 Ultra (free)NVIDIA · 442.4B tokens2.5%—
10GLM 5.3Z.ai · 421.8B tokens2.4%▲2

New Models

4
DeepSeek Pro LatestDeepSeek · 1.0M ctx · 09/14$0.58 / $1.74
DeepSeek Flash LatestDeepSeek · 1.0M ctx · 09/14$0.15 / $0.60
Schematron V2 TurboInference.net · 128.0k ctx · 09/12$0.03 / $0.15
Schematron V2 SmallInference.net · 128.0k ctx · 09/12$0.05 / $0.23

Deals

24 h · 4 deals

Bloomberg reports OpenAI is in early talks with investors about a funding round that woul…

Bloomberg ·
$1.2T+
funding round

Model ML, which sells AI software to investment banks and financial firms, is in talks to…

Bloomberg ·
$100M