Thinking Machines drops Inkling 975B open-weights; Apple-Qwen approved for China; Codex Micro keyboard ships

Mira Murati's Thinking Machines released Inkling — a 975B-parameter open-weights model trained to understand video and audio (1,024 HN). Apple Intelligence was approved for launch in China with Alibaba's Qwen. OpenAI's first branded hardware turns out to be the Codex Micro — a $230 light-up keyboard for monitoring agentic threads, not the speaker. Suno was hacked, exposing that its AI music generator scraped YouTube, Genius, and Deezer at scale. xAI is now suing a Grok user over CSAM — defensive countermove after last week's own suit. Microsoft is reportedly training its salespeople to talk down OpenAI and Anthropic. Anthropic and Blackstone bet on services with Ode. DeepMind + Isomorphic Labs published a joint bioresilience approach. SpaceX shares fell to $135 as the euphoric post-IPO high fades ahead of Starship.

7 papers 19 news 9 sources ← Latest

News

21 items

Thinking Machines drops Inkling 975B; the open-vs-frontier fight compounds

Thinking Machines released Inkling, a 975B-parameter open-weights model trained on video + audio (1,024 HN). Wired and TC frame it as the first public proof point from Murati's team. The Siegel Endowment published a widely-shared PDF (209 HN) arguing governments, companies, and nonprofits should invest in free, open-source AI. Ring-Zero scales zero-RL to a trillion parameters for emergent reasoning.

News Hacker News

Inkling: Our Open-Weights Model

Thinking Machines ships Inkling — 975B-parameter open-weights model trained on video and audio (1,024 HN points).

params 975B

Why it matters

First public model from Mira Murati's Thinking Machines after 18 months of infra work.
975B open-weights across video + audio directly compounds Gemma 4, GLM 5.2, and LongCat 2.0.
Materially strengthens Delangue's 'real race isn't at the frontier' thesis with a US-founded open lab.
Sets a new comp for open-weights valuation and the enterprise 'own vs rent' conversation.

Source →

News TechCrunch AI

Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling

TC: Inkling is Thinking Machines' first public proof point — a bet against one-size-fits-all AI.

products open-weights market

Source →

News Wired AI

Thinking Machines Lab Drops Its First Model

Wired: Inkling is 975B, open source, and trained to understand video and audio.

products open-weights multimodal

Source →

News Hacker News

Governments, companies, nonprofits should invest in free, open source AI

Siegel Endowment PDF (209 HN points): open-source AI should be funded by governments, companies, and nonprofits.

policy open-weights community

Source →

OpenAI's real hardware is a keyboard; Apple Intelligence gets China via Qwen

OpenAI shipped the Codex Micro — a $230 light-up keyboard for monitoring agentic threads — not the speaker. Apple Intelligence was approved for launch in China with Alibaba's Qwen AI. Microsoft is reportedly training salespeople to talk down OpenAI and Anthropic to sell its in-house models.

News TechCrunch AI

Amid hardware legal battle, OpenAI releases a $230 keyboard for Codex

TC: OpenAI ships a $230 Codex-branded keyboard during the Apple hardware suit.

price $230

Why it matters

First OpenAI-branded hardware product, priced at $230 — during the Apple trade-secret suit.
Reframes yesterday's 'moving speaker' report as a separate future product, not this launch.
Sets a physical-product precedent OpenAI can point to as it fights Apple's IP claims.
Directly relevant to the Codex-hours-long agent workflow and ChatGPT-Work rollout.

products code agents

Source →

News Ars Technica AI

OpenAI's first branded hardware is... a light-up keyboard?

Ars: Codex Micro is a light-up keyboard for monitoring multiple agentic threads at a glance.

products code agents

Source →

News The Verge AI

OpenAI finally launches hardware… for Codex

The Verge: OpenAI's first hardware isn't the mysterious device — it's Codex Micro.

products code

Source →

News TechCrunch AI

Apple Intelligence approved for launch in China with Alibaba's Qwen AI

TC: Apple Intelligence is approved for China with Alibaba's Qwen as the local model provider.

Why it matters

First major US-frontier consumer-AI product officially launching in China with a Chinese-lab backbone.
Direct answer to the Alibaba-Claude-Code ban and the Anthropic-Alibaba fight.
Materially affects the sovereign-AI framing — Apple is choosing multi-provider by geography.
Sets the template Meta/Google will now be pressured to match for China distribution.

policy market products

Source →

News TechCrunch AI

Microsoft is reportedly training salespeople to talk down OpenAI and Anthropic

TC: Microsoft is training its enterprise salespeople to position OpenAI and Anthropic as more expensive and less efficient than MSFT's in-house models.

Why it matters

Direct MSFT-vs-OpenAI/Anthropic go-to-market shift after Nadella's warning to AI users.
Concrete operator-side evidence for the OpenAI-MSFT breakup chatter.
Materially affects the Copilot-preferred-model story from last week — MSFT playing both sides.

market products policy

Source →

Suno scraped YouTube et al; xAI turns predator suit around; HF security incident

TC and The Verge: Suno was hacked, exposing that the AI music generator scraped YouTube, Genius, and Deezer. xAI turned the CSAM story around by suing a South Carolina Grok user who allegedly generated CSAM deepfakes. Hugging Face disclosed a July 2026 security incident on its own blog.

News TechCrunch AI

Hack suggests AI music generator Suno scraped YouTube for training data

TC: A hack exposed Suno source code showing it scraped YouTube for training data.

Why it matters

Direct exhibit for the NYT-vs-OpenAI copyright fight and the Cloudflare AI-crawl payment deadline.
Materially affects Suno's Spark artist-incubator program and the TIDAL demonetize-AI-music policy.
Sets a public precedent for how AI-training-data provenance will be treated in future rounds of suits.
Compounds Google's fresh Hachette/Cengage/Elsevier copyright suit from yesterday.

policy regulation safety market

Source →

News The Verge AI

Suno snatched millions of songs from YouTube, Genius, and Deezer

The Verge: Suno data from the hack showed training on millions of songs from YouTube, Genius, and Deezer.

policy regulation safety

Source →

News The Verge AI

xAI sues a man for using Grok to generate CSAM 'deepfakes'

The Verge: xAI is suing a South Carolina man alleged to have used Grok to generate CSAM deepfakes.

Why it matters

Defensive countermove after xAI itself was sued over Grok CSAM last week.
Sets a legal-strategy template: reframe the CSAM story as a user-abuse suit rather than product failure.
Materially affects the xAI-Grok safety-narrative going into IPO conversations.

policy regulation safety

Source →

News Hugging Face

Security incident disclosure — July 2026

Hugging Face publishes a formal July 2026 security-incident disclosure.

Why it matters

First formal HF security-incident disclosure in the model-hub-and-community layer.
Materially affects trust posture for open-weights distribution pipelines.
Compounds the vendor-agent transparency incidents from last week.

safety open-weights community

Source →

Anthropic + Blackstone launch Ode; Emergent unicorn; Rime $24M

Anthropic and Blackstone launched Ode — a bet that AI-services (implementation) is the next trillion-dollar business, not just models. Indian AI-coding startup Emergent hit unicorn status with a $130M Series C. Voice-AI startup Rime raised $24M Series A after handling 100M+ calls a month. SpaceX shares fell to $135 IPO price ahead of Starship.

News TechCrunch AI

Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just models

TC: Anthropic-backed Ode launches with Blackstone — betting AI-implementation is a trillion-dollar business.

Why it matters

First frontier-lab / private-equity JV around AI-services implementation.
Direct extension of the Amazon/MSFT/Anthropic FDE motion from earlier in the month.
Materially affects the Big Consulting narrative — Accenture and Deloitte squarely targeted.

funding products market

Source →

News TechCrunch AI

Indian AI coding startup Emergent becomes a unicorn with $130M Series C

TC: Emergent hits unicorn after $130M Series C with $120M annualized run rate — under 18 months from launch.

raise $130MARR $120M

Why it matters

Third vibe-coding-adjacent unicorn this cycle after SpaceX-Cursor and Lovable.
Direct India-side signal alongside OpenAI's India hire and Anthropic's India pricing.

funding code products

Source →

News TechCrunch AI

Rime picks up $24M Series A to help enterprises field customer calls

TC: Rime raised $24M Series A; handling 100M+ enterprise customer calls per month.

funding products multimodal

Source →

News TechCrunch AI

SpaceX falls to $135 IPO price ahead of Starship launch

TC: SpaceX shares fell to the $135 IPO price — the euphoric post-IPO high is fully gone.

share price $135 (IPO)

Why it matters

First public-market signal cooling the SpaceX-as-AI-stack narrative.
Relevant to the OpenAI/Anthropic/SpaceX 'bigger than 25 years of tech exits' framing.
Materially affects the Cursor-inside-SpaceX and Grok Build competitive story.

market funding

Source →

OpenAI's 'reverse federalism'; DeepMind bioresilience; GPT-Red; Vint Cerf's agent IDs

OpenAI published its 'reverse federalism' AI safety framework — state laws building up to federal. DeepMind + Isomorphic Labs shared their joint bioresilience approach. MIT Tech Review reveals GPT-Red, OpenAI's LLM super-hacker built to make its own models safer. Vint Cerf is working on a standard for identifying AI agents in the wild.

News OpenAI

The US is advancing AI safety through state and federal action

OpenAI outlines a 'reverse federalism' approach — state laws building toward federal AI governance.

Why it matters

First OpenAI public policy framework for state-vs-federal AI governance.
Direct answer to the NY data-center moratorium and Anthropic-Newsom half-price deal.
Sets the reference OpenAI will use to frame state-level AI politics through 2027.

policy regulation safety

Source →

News Google DeepMind

Our approach to bioresilience

DeepMind + Isomorphic Labs publish a joint bioresilience-and-AI approach.

Why it matters

First joint DeepMind + Isomorphic Labs public bioresilience framework.
Direct answer to Anthropic Claude Science's drug-development ambition.
Sets the biosecurity-adjacent framework OpenAI's new Miles Wang venture will need to address.

policy safety science

Source →

News MIT Technology Review

Meet GPT-Red: an LLM super-hacker OpenAI built to make its models safer

MIT TR: GPT-Red is OpenAI's internal LLM super-hacker used as a red-team sparring partner.

Why it matters

First public disclosure of OpenAI's dedicated internal red-team LLM.
Sets a reference for how frontier labs will do continuous adversarial testing.
Directly relevant to the CVE-spike-around-Mythos-Preview data and the GitLost/HalluSquatting incidents.

safety products evaluation

Source →

News TechCrunch AI

Vint Cerf is working on a plan to unleash AI agents on the open internet

TC: Vint Cerf is designing a standard for identifying AI agents in the wild — TCP/IP for agents.

Why it matters

First TCP/IP-scale standards attempt for identifying AI agents on the open internet.
Direct answer to the CUGA/MCP/X-MCP wave — the field lacks an identity layer.
Materially affects HTTP referer, robots.txt, and Cloudflare's payments deadline story.

policy agents infrastructure

Source →

Papers

1 item

Thinking Machines drops Inkling 975B; the open-vs-frontier fight compounds

Paper Hugging Face

Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning

Scales zero-RL to a trillion parameters for emergent reasoning.

reinforcement-learning reasoning training

Source → Arc

Also today

News · TechCrunch AI Microsoft patches record number of security vulnerabilities, citing its use of AI — TC: Microsoft Patch Tuesday resolved a record 570+ vulnerabilities, MS citing AI as the enabling tool.
News · TechCrunch AI Applied Computing wants to give oil and gas operators an AI model for the entire plant — TC: Applied Computing raised $20M Series A for a foundation AI model targeting oil-and-gas plants.
News · TechCrunch AI Whatnot acquires Shaped to power real-time live shopping recommendations — TC: Whatnot acquired ML startup Shaped for real-time live-shopping recommendations.
News · TechCrunch AI Reelful's AI turns your camera roll into short-form videos for social media — TC: Reelful automates short-form social video from your camera roll.
News · TechCrunch AI Inside Ode with Anthropic, the startup betting AI services are the future of enterprise — TC video interview inside Ode — the Anthropic-backed AI-services company.
News · The Verge AI COMPUTER COPS: Inside the big business of selling AI to the police — The Verge feature: inside the business of selling AI to US police departments.
News · The Verge AI AI slop movies are the new direct-to-video cash grabs — The Verge: AI slop movies are landing in theaters this weekend as direct-to-video cash grabs.
News · Wired AI An Inventor of Apple's FaceID Wants to Analyze Your Brain's Health With AI — Wired: Ex-Apple FaceID veteran Gidi Littwin's Hemispheric ships AI brain-health scans.
News · Wired AI Please Stop Making Me Opt Out of AI — Wired opinion: enough with automatically-enabled AI features and opt-out toggles.
News · Wired AI AI Isn't Smarter Than a Baby — Yet — Wired: Baby learning is where the next AI advances may come from.
News · Hugging Face What building Shippy taught us about building agents — Allen AI on lessons learned from building Shippy as an agent.
News · Hugging Face Model Routing Is Simple. Until It Isn't. — IBM Research on the practical complexity of model routing.
News · Hacker News LLM Networking with MikroTik — Practitioner post (95 HN) on LLM-driven MikroTik network config.
Paper · Hugging Face Self-Improvements in Modern Agentic Systems: A Survey — Survey of self-improvement techniques in modern agentic systems.
Paper · Hugging Face PalmClaw: A Native On-Device Agent Framework for Mobile Phones — Native on-device agent framework for mobile phones.
Paper · Hugging Face Boogu-Image-0.1: Boosting Open-Source Unified Multimodal Understanding and Generation — Boogu-Image-0.1 pushes open-source unified multimodal understanding + generation.
Paper · Hugging Face Tracing Agentic Failure from the Flow of Success — Tracing agentic failure modes from success trajectories.
Paper · Hugging Face Harness Handbook: Making Evolving Agent Harnesses Readable, Navigable, and Editable — Handbook for making evolving agent harnesses readable and editable.