Tech Express | September 9, 2026 Apple's Foldable 'iPhone Duo' Reportedly Starts at $2,000

1. Apple's foldable 'iPhone Duo' will reportedly start at $2,000

Source: The Verge

With just hours left before the first iPhone launch event since John Ternus took over as Apple's CEO, Bloomberg reporter Mark Gurman says the company's long-rumored foldable phone will be called the "iPhone Duo."

According to the report, the device will start at $2,000 for a 256GB model, or around $3,000 for up to 2TB of storage. Shipping is said to begin as soon as October in two colors: white and dark blue.

The rumored foldable had mostly been referred to as the "iPhone Ultra." The Duo name was previously used by Google for a messaging app and by Microsoft for its Surface Duo devices. The starting price would put it above Google's Pixel 11 Pro Fold and Samsung's Galaxy Z Fold 8 ($1,899), but below the Galaxy Z Fold 8 Ultra ($2,099.99).

Source

2. OpenAI's GPT-6 Astra reclaims the top spot as rivalry with Anthropic heats up

Source: Huxiu

On September 3, OpenAI released GPT-6 Astra, calling it "the world's most intelligent model." In tests covering software engineering, computer use and professional work, Astra significantly outperformed the previous GPT-5.6 Sol and led Claude on several metrics.

It scored 97.6% on FrontierMath Tier 4, up to 99.9% on ARC-AGI-3, and 100% on ExploitBench. In more cautious independent rankings, as of September 7, Astra tied Claude Fable 5.1 for first place with 53 points on the Artificial Analysis index.

OpenAI positioned Astra as a model "for the hardest end-to-end work," emphasizing coding, computer use and complex multi-step tasks. The launch was widely viewed through the lens of OpenAI's rivalry with Anthropic, whose valuation has reached $965 billion, surpassing OpenAI's $852 billion.

Source

3. OpenAI releases ChatGPT Images 2.5 with up to 50% lower latency

Source: 9to5Mac

OpenAI has released ChatGPT Images 2.5, an upgraded image generation model touting "sharper details, faster generation, more precise editing, and better tools for creating and sharing."

Compared with Images 2, released in April, the new version cuts image generation latency by up to 50%, produces more natural lighting and richer textures, better preserves subjects in reference photos, and follows editing instructions more reliably across multiple turns.

The update also adds a "Sketch" feature for drawing directly in ChatGPT as a reference, plus templates and prompt sharing. For developers, OpenAI is offering two API versions: GPT-Image-2.5 Flare (the default) and GPT-Image-2.5 Sunburst for premium workflows.

Source

4. Former OpenAI researcher Tian Yonglong joins Tencent to lead Hunyuan multimodal

Source: MyDrivers

According to a report by The Paper, former OpenAI researcher Tian Yonglong has joined Tencent's CEO/President Office as head of Hunyuan multimodal, reporting directly to Tencent's large-model lead Yao Shunyu.

His public resume shows he studied at Tsinghua University, then at the Chinese University of Hong Kong's MMLab, before earning a PhD at MIT. After his doctorate he worked at Google Research, Google DeepMind and OpenAI. He officially joined Tencent in July 2026 to lead the Hunyuan multimodal direction.

On July 24, Tencent merged its Hunyuan multimodal division with its large language model division into a foundation model unit managed by Yao Shunyu, who also previously worked at OpenAI after earning a PhD at Princeton.

Source

5. Samsung opens AI semiconductor R&D base in Yokohama, teaming with Japanese firms

Source: Nikkei

Samsung Electronics held an opening ceremony on September 8 in Yokohama for an AI semiconductor R&D base, teaming up with Japanese firms that hold the world's largest share in semiconductor assembly equipment and materials.

The facility, called APL (Advanced Package Lab), sits in the Minato Mirai area and spans about 6,600 square meters with a cleanroom. Total investment is 40 billion yen, of which 22.5 billion yen comes from Japanese government support, and 95 Japanese and Korean researchers will work on materials evaluation and mass-production technology.

Samsung aims to bring in Japan's world-leading back-end process technology. As AI data centers drive surging demand for high-bandwidth memory (HBM), assembly has become a new bottleneck; Samsung says it already works with more than 50 Japanese companies on AI semiconductor back-end processes.

Source

6. JD Cloud and Moore Threads to build 100,000-GPU cluster

Source: MyDrivers

At the 2026 JD Global Tech Discovery Conference on September 9, JD Cloud announced plans to build a 100,000-card full-function GPU cluster, using Moore Threads GPUs as its compute foundation and targeting large-model training, inference and embodied intelligence.

This marks the first time domestic Chinese GPUs have entered the core 100,000-card intelligent computing cluster of a major AI cloud firm. JD Cloud president Cao Peng said the partnership was driven by Moore Threads' rare capabilities in domestic training-grade GPUs and mature large-cluster delivery experience.

Moore Threads founder, chairman and CEO Zhang Jianzhong called full-function GPUs the preferred choice for building premium domestic computing infrastructure, with the two sides collaborating across chips, cloud platforms and model training to power JD's JoyAI models.

Source

7. OPPO opens Find X10 reservations; Pro Max debuts triple 200MP cameras

Source: MyDrivers

On September 9, OPPO announced that the Find X10 series is now open for reservations across all channels, including the Find X10 and the Find X10 Pro Max. The Find X10 is positioned as a "one-step imaging flagship" with a dual 200-megapixel camera system.

The Find X10 Pro Max debuts a triple 200-megapixel camera system co-developed with Hasselblad, with the main, ultra-wide and telephoto cameras all using 200-megapixel sensors. The Hasselblad 200MP main camera debuts Samsung's HPC sensor and DeepPix dynamic technology, reaching up to 17EV of dynamic range, with CIPA 6.5-level stabilization and an f/1.5 aperture.

The series also packs an 8,000mAh "glacier" battery rated for five years of durability, a new-generation 1-nit eye-care display, and debuts ColorOS 17.

Source

8. Asteroid 2026 RW1 spotted just hours before hitting Earth

Source: BBC Sky at Night Magazine

A tiny asteroid just 80cm (2.6 feet) wide was detected only hours before it entered Earth's atmosphere — the 13th known impacting asteroid ever spotted before it hit Earth.

On Sunday, September 6, automated telescopes at Mount Lemmon Observatory in Arizona picked up the object, now officially designated 2026 RW1. Six hours after discovery, it burned up over the Indian Ocean off the northwest coast of Australia, causing no damage.

After detection, the European Space Agency's Meerkat automated warning system studied the asteroid and predicted its trajectory and impact time. Planetary defence experts see the detection as a success and evidence that early-warning systems for small objects are improving.

Source

9. US agencies accuse six Chinese firms of distilling American AI models

Source: Zaobao

The US National Security Agency, FBI and Cybersecurity and Infrastructure Security Agency issued a joint statement on September 8, accusing six Chinese AI companies — including DeepSeek and Moonshot AI — of systematically extracting proprietary knowledge from US AI models.

The statement said the firms have used "distillation" techniques at "industrial scale" since at least 2024, naming DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Zhipu, with US models from Anthropic, OpenAI, Google and SpaceX involved.

The US agencies described "aggressive, malicious and targeted distillation activities" and urged American developers to adjust model responses and share intelligence. Neither the Chinese embassy nor the six companies immediately responded to requests for comment.

Source

10. Zero-click WeChat voice-call worm hijacks accounts without interaction

Source: MyDrivers

Security research firm Calif Research has published a zero-click worm attack demonstration called WeWorm, the first attack scheme that automatically spreads between iOS and Android via WeChat voice calls, hijacking accounts without any action from the victim.

In the demo, a Pixel 10a called an iPhone 17e over WeChat and hijacked it while the phone was still ringing; the controlled iPhone then called a third phone to repeat the attack. The whole process takes just seconds, and victims hear nothing.

The flaw is a memory-corruption issue in WeChat's VoIP protocol stack. Calif Research found the vulnerability with AI assistance, and Tencent patched it on August 21 with Android 8.0.77 and iOS 8.0.76.

Source