Triweb AI
Back to AI Insights
Latest16 Jul 2026Triweb AI4 min read

Thinking Machines Unveils Inkling, OpenAI Ships a Codex Keyboard, and Apple Brings AI to China

Thinking Machines released Inkling, its first open-weights model with 975 billion parameters. OpenAI launched a $230 keyboard for Codex amid a legal dispute with Apple. Apple secured approval to bring AI features to China through a partnership with Alibaba's Qwen.

Thinking Machines Unveils Inkling, OpenAI Ships a Codex Keyboard, and Apple Brings AI to China

OpenAI released a custom keyboard for its coding assistant Codex while Apple secured approval to launch AI features in China with Alibaba's Qwen models. Thinking Machines also stepped into the spotlight with Inkling, its first open-weights model.

Thinking Machines Inkling Aims at Enterprise AI With Open Weights

Thinking Machines released Inkling, a 975-billion parameter mixture-of-experts model that marks the company's first public AI system after 18 months of quiet infrastructure building. The model activates only 41 billion parameters per task, a design choice that keeps inference costs manageable while maintaining broad capability.

The company is positioning Inkling primarily as an enterprise research platform rather than a finished product. It supports calibrated confidence outputs, meaning the model can flag uncertainty instead of guessing, and lets users adjust "thinking effort" to trade accuracy for speed. Thinking Machines says it will release the model openly for researchers and developers to inspect and build on.

Inkling was trained entirely on Nvidia hardware under a partnership the companies struck in March. Thinking Machines does not claim Inkling is the strongest model available. Its blog post states plainly that the model is "not the strongest overall model available today, open or closed." The focus instead is on efficiency, transparency, and giving enterprises more control over the models they depend on.

The approach is a deliberate contrast to the general-purpose chatbot strategy used by OpenAI, Anthropic, and Google. Thinking Machines is betting that enterprises want models they can customize and trust, not just black-box APIs.

OpenAI Launches a $230 Keyboard for Codex

OpenAI released a physical keyboard designed specifically for Codex, its AI-powered coding assistant. The keyboard includes a dedicated row of programmable keys for code completion, refactoring, and agent interaction workflows. It also features RGB backlighting that changes based on Codex's status.

The launch comes during an ongoing legal dispute with Apple, which has accused OpenAI of stealing hardware trade secrets related to keyboard and input technology. OpenAI denies the allegations but the timing of the announcement has drawn scrutiny. Priced at $230, the keyboard is aimed at professional developers who use Codex daily.

The hardware release signals OpenAI's broader push beyond software subscriptions. The company has been investing in physical products as a new revenue stream, particularly for its developer tools.

Apple Intelligence Gets China Approval With Alibaba's Qwen

Apple received regulatory approval to launch Apple Intelligence features in China through a partnership with Alibaba's Qwen AI division. The deal was rumored for months and marks Apple's first major AI partnership in the Chinese market.

The approval allows Apple to bring AI-powered features like enhanced Siri, image generation, and text summarization to iPhone and Mac users in China. Apple chose to work with a local partner rather than import models from US-based AI companies, a decision shaped by China's data sovereignty regulations.

The partnership also gives Alibaba's Qwen technology a major distribution channel. Qwen has been competing with models from Baidu, ByteDance, and others for dominance in China's rapidly growing AI market. Being embedded in Apple's ecosystem provides a significant credibility and reach advantage.

Suno Hack Reveals YouTube Scraping for Training Data

A security breach at AI music generation company Suno exposed source code showing the company scraped decades of audio from YouTube to train its models. The hacker used an employee's credentials to access the codebase.

The revelation adds fuel to the ongoing debate about training data provenance in generative AI. Suno's technology can produce convincing songs in any style based on text prompts. If the scraping violated YouTube's terms of service or copyright law, the company could face legal exposure. Suno has not commented on the breach or the scraping claims.

This article was written by Triweb AI's editorial team based on analysis of today's leading AI news sources.

Share this article

Get AI insights delivered weekly

Like what you read? Join the newsletter and receive the latest AI news, analysis, and practical insights straight to your inbox every week - written by Triweb AI.

Free AI Implementation Starter Kit