On October 28, 2026, Iluvatar CoreX conducted its first live demonstration of the general-purpose GPU Tiange 300 and the edge computing terminal TY1200, and previewed the next-generation edge product TY2000. According to a report by Zhidongxi, the Tiange 300 is equipped with 144GB of memory, delivering Attention performance 1.05 times that of the H200 under a 64K long context, and MoE throughput 1.15 times that of the H200 on the GLM-5.3 model; the TY1200 offers 300 TOPS of dense INT8 compute, with measured TPS surpassing NVIDIA's Thor T5000. The Tiange 300 IXL144 supernode integrates 144 GPUs in a single rack, aggregating approximately 10.1TB of memory, and reduces the cost per million effective tokens by about 40% compared with mainstream international solutions. Currently, this event has only the single report described above, with no subsequent developments.
Singapore-based manufacturer Brilliant Labs released the Halo AI glasses on local time October 27, priced at $399. The device features a full-color Micro-OLED display, is equipped with the self-developed AI assistant Noa, supports real-time voice interaction and long-term memory, and is built around three experiences: Ask, Make, and Play, among which the Narrative system can turn real-world activities into interactive games that can be adjusted in real time. Both hardware and software adopt an open-source design, with on-device AI processing at the core; the device weighs slightly more than 40 grams, and the company claims a maximum battery life of 14 hours. Separately, according to a Unite.AI report, Noa and miniapps have daily usage limits, with Noa Plus at $9.99 per month and Noa Pro at $39.99 per month; the company also announced an investment from Goodwater Capital, without disclosing the amount.
Singapore-based manufacturer Brilliant Labs released the Halo AI glasses on local time October 27, priced at $399. The device features a full-color Micro-OLED display, is equipped with the self-developed AI assistant Noa, supports real-time voice interaction and long-term memory, and is built around three experiences: Ask, Make, and Play, among which the Narrative system can turn real-world activities into interactive games that can be adjusted in real time. Both hardware and software adopt an o
On October 29, 2026, ISHO Today's Relaxation released the ISHO Ring smart ring and the ISHO iFlow smart temperature control system. According to a report by Zhidongxi, the ISHO Ring is priced at 1,799 yuan, with a charging case version at 2,199 yuan; it is the world's first to feature the Apollo 330B Plus chip, can run AI monitoring models on-device, and offers 6-10 days of battery life. The iFlow starts at 9,999 yuan and is equipped with the MediaTek Genio 420 platform. A supplementary report by Shangguan News on October 30 noted that the ISHO Ring comes with 32MB of storage and can continuously monitor 18 data metrics; the iFlow has 7.2 TOPS of on-device computing power and uses TEC semiconductor temperature control modules to achieve independent dual-zone temperature adjustment for two people. A personal agent app and the ISHO physiological signal large model were also launched at the same time. Wang Teng said the company aims to build a globally leading AI sleep technology ecosystem within 5-10 years, and the team has already grown to more than 100 people.
On October 29, 2026, ISHO Today's Relaxation released the ISHO Ring smart ring and the ISHO iFlow smart temperature control system. According to a report by Zhidongxi, the ISHO Ring is priced at 1,799 yuan, with a charging case version at 2,199 yuan; it is the world's first to feature the Apollo 330B Plus chip, can run AI monitoring models on-device, and offers 6-10 days of battery life. The iFlow starts at 9,999 yuan and is equipped with the MediaTek Genio 420 platform. A supplementary report b
On October 29, 2026, Diffusion Intelligence released the diffusion language model Clover-35B-A3B-Preview. According to QbitAI, the model uses a MoE architecture with 35 billion total parameters and about 3 billion activated per token, and supports a thinking mode; in the 49 tasks of the web Agent benchmark WindTunnel, its median execution time was 1.52 seconds, lower than the 3.21 seconds of the Jev+Mercury 2.5 combination, with success rates of 98.6% and 95.9% respectively; on code benchmarks it scored 73.6 on SWE-bench Verified and 51.7 on Terminal-Bench 2.1. The team said it will subsequently launch a multimodal version and advance on-device adaptation. On October 30, IT Home · Artificial Intelligence reported that the artificial intelligence lab DiffuSpace released Clover-35B-A3B-Preview, the first model in the Clover series, which uses a MoE architecture and is open-sourced under the Apache license, generating results through multiple rounds of iteration; DiffuSpace said it outperforms the same-scale autoregressive model Qwen3.6-35B-A3B on multiple benchmarks while maintaining faster generation speed, and is especially good at agentic coding and tool-use tasks.
On October 29, 2026, Diffusion Intelligence released the diffusion language model Clover-35B-A3B-Preview. According to QbitAI, the model uses a MoE architecture with 35 billion total parameters and about 3 billion activated per token, and supports a thinking mode; in the 49 tasks of the web Agent benchmark WindTunnel, its median execution time was 1.52 seconds, lower than the 3.21 seconds of the Jev+Mercury 2.5 combination, with success rates of 98.6% and 95.9% respectively; on code benchmarks i
On October 28, 2026, Iluvatar CoreX opened the Tiange 300 live machine demonstration to the media for the first time at a media open day. Phoenix Technology reported that the Tiange 300 outperformed the H200 overall in comparative tests involving large language models such as GLM-5.3 as well as text-to-image and text-to-video scenarios; in an 8-card full-machine FP8-precision demonstration based on Qwen3-32B, the effective throughput per card reached 1.33 times that of the H100. A subsequent report by Shangguan News stated that in an on-site random-question comparison, the Tiange 300 took 95.6 seconds while NVIDIA's H200 took 112 seconds, and noted that third-party institution testing in July this year showed that under PD-disaggregated deployment, the Tiange 300's converted single-card end-to-end throughput reached 1.05 times that of the H200. The same event also showcased the AFD disaggregation solution, the IXL144 super node and the Tongyang TY1200 edge-side terminal, and previewed the next-generation edge-side product TY2000 with a nominal computing power of 600T.
On October 28, 2026, Iluvatar CoreX opened the Tiange 300 live machine demonstration to the media for the first time at a media open day. Phoenix Technology reported that the Tiange 300 outperformed the H200 overall in comparative tests involving large language models such as GLM-5.3 as well as text-to-image and text-to-video scenarios; in an 8-card full-machine FP8-precision demonstration based on Qwen3-32B, the effective throughput per card reached 1.33 times that of the H100. A subsequent rep
On October 27, 2026, Apple released a new iMac equipped with the M6 chip. According to Apple Newsroom, the new machine features up to a 12-core CPU and a 12-core GPU (each core containing Neural Accelerators), with on-device AI performance up to 5 times faster than M4 and up to 12 times faster than M1; it uses a Dual 16-core Neural Engine, memory bandwidth up to 170GB/s, supports Wi-Fi 7, Bluetooth 6, and up to 2.5 Gigabit Ethernet, and comes with a 24-inch 4.5K Retina display and macOS 27, offered in seven color options. Pre-orders open on October 28, with sales starting on November 6. The next day, Jiwei.com reported that the new iMac starts at $1,499, and that its on-device AI performance is up to 4 times faster than models equipped with the M4 chip, which differs from Apple Newsroom's statement of up to 5 times faster than M4; the report also said that the United States and 34 other countries and regions can pre-order from October 28, with official sales beginning on November 6.
On October 27, 2026, Apple released a new iMac equipped with the M6 chip. According to Apple Newsroom, the new machine features up to a 12-core CPU and a 12-core GPU (each core containing Neural Accelerators), with on-device AI performance up to 5 times faster than M4 and up to 12 times faster than M1; it uses a Dual 16-core Neural Engine, memory bandwidth up to 170GB/s, supports Wi-Fi 7, Bluetooth 6, and up to 2.5 Gigabit Ethernet, and comes with a 24-inch 4.5K Retina display and macOS 27, offe
On October 27, 2026, Apple released a new 14-inch MacBook Pro equipped with the M6 chip. According to information published on Apple Newsroom, the new machine features a 12-core CPU, a 12-core GPU, and a dual 16-core Neural Engine, with battery life of up to 24 hours. Compared with the M5, graphics performance is up to 30% faster, and on-device AI workloads such as LLM prompt processing are up to 25% faster; upgrading from the M1 delivers graphics performance up to 11 times faster and AI performance up to 13 times faster. The new machine comes standard with 16GB of unified memory, configurable up to 32GB, supports Wi-Fi 7 and Bluetooth 6, with pre-orders opening on October 28 and sales starting on November 6. There are currently no further developments.
On October 27, 2026, Apple released a new 14-inch MacBook Pro equipped with the M6 chip. According to information published on Apple Newsroom, the new machine features a 12-core CPU, a 12-core GPU, and a dual 16-core Neural Engine, with battery life of up to 24 hours. Compared with the M5, graphics performance is up to 30% faster, and on-device AI workloads such as LLM prompt processing are up to 25% faster; upgrading from the M1 delivers graphics performance up to 11 times faster and AI perform
Bold Security co-founder and CEO Nati Hazut put forward this judgment: AI shifts sensitive work to endpoints, and endpoint security needs to cover the data interactions of employees and autonomous agents. According to an interview published by Unite.AI on October 24, 2026, Hazut introduced that the company uses small language models running directly on devices to identify sensitive data and how it is used, covering browsers, desktop applications, and AI agents. He pointed out that desktop AI applications and agents bypass the visibility of network detection, CASB, and DSPM, becoming new channels for data exfiltration; endpoint-side solutions can intervene in real time before data leaves the enterprise's control. Currently, this judgment is presented by this single interview report, and there are no subsequent factual developments.
Bold Security co-founder and CEO Nati Hazut put forward this judgment: AI shifts sensitive work to endpoints, and endpoint security needs to cover the data interactions of employees and autonomous agents. According to an interview published by Unite.AI on October 24, 2026, Hazut introduced that the company uses small language models running directly on devices to identify sensitive data and how it is used, covering browsers, desktop applications, and AI agents. He pointed out that desktop AI app
The team led by Dai Qionghai at Tsinghua University and the team led by Dai Pu at Chinese PLA General Hospital have developed an ultra-light wireless mesoscopic microscope, Airscope. Weighing only 1 gram, the device can record neural activity in the mouse cerebral cortex within a 6-millimeter field of view at a single-cell resolution of 4 to 6 micrometers, capturing more than 8,500 neurons simultaneously in a single mouse—25 times that of existing devices of the same type. The device uses only 4 aspheric lenses and achieves on-device compression through a 2-megapixel image sensor, enabling wireless transmission of megapixel-level images and freeing it from cable constraints. The results were published online in Nature on the 19th. The team has used it to record cortical activity in multiple mice across a 5-square-meter natural habitat and a large maze with 32 endpoints.
The team led by Dai Qionghai at Tsinghua University and the team led by Dai Pu at Chinese PLA General Hospital have developed an ultra-light wireless mesoscopic microscope, Airscope. Weighing only 1 gram, the device can record neural activity in the mouse cerebral cortex within a 6-millimeter field of view at a single-cell resolution of 4 to 6 micrometers, capturing more than 8,500 neurons simultaneously in a single mouse—25 times that of existing devices of the same type. The device uses only 4
On October 24, 2026, OpenBMB released research on a device-cloud collaborative long-horizon agent execution system. This research, conducted by ModelBest in collaboration with Tsinghua University, uses EC2-Bench to evaluate four device-cloud collaboration strategies on 128 executable tasks, simultaneously measuring task completion, resource cost, and privacy exposure. The results show that with on-device Qwen3.5-9B plus cloud-based GLM-5.3, the best collaboration score is 60.46, about 73% of the pure-cloud score of 82.70, at a cost of only $8.73, about 15% of the pure-cloud cost of $56.42. The research also found that weak on-device models are suited to having an external router let the cloud execute, with the 2B model scoring 39.97 versus 15.86; strong on-device models are suited to the cloud providing only suggestions, with the 9B model scoring 60.46. The research has now been made public, with no subsequent progress yet.
On October 24, 2026, OpenBMB released research on a device-cloud collaborative long-horizon agent execution system. This research, conducted by ModelBest in collaboration with Tsinghua University, uses EC2-Bench to evaluate four device-cloud collaboration strategies on 128 executable tasks, simultaneously measuring task completion, resource cost, and privacy exposure. The results show that with on-device Qwen3.5-9B plus cloud-based GLM-5.3, the best collaboration score is 60.46, about 73% of the
On October 22, 2026, iFLYTEK released the iFlytek Spark X2.6 device-cloud integrated large model, comprising the cloud-based 293-A30B model and the already open-sourced on-device X2.6-4B model, both of which completed capability upgrades based on fully domestic computing power. According to Zhi Dongxi reports, the cloud model supports a one-million-token context, with an input price of 0.9 yuan per million tokens and an output price of 3.6 yuan per million tokens. At the same event, iFLYTEK also released the SparkAudio 1.0 speech foundation model, the Robot Super Brain Platform V3.0, the WinWork digital employee platform, and hardware such as the iFlytek AI recording ring. On the same day, iFLYTEK Chairman Liu Qingfeng said at the main forum of the 2026 Global 1024 Developer Festival that domestic computing power still lags about two generations behind the global level, but will soon narrow to one generation or half a generation, and there will certainly be a day when it runs neck and neck or even leads.
On October 22, 2026, iFLYTEK released the iFlytek Spark X2.6 device-cloud integrated large model, comprising the cloud-based 293-A30B model and the already open-sourced on-device X2.6-4B model, both of which completed capability upgrades based on fully domestic computing power. According to Zhi Dongxi reports, the cloud model supports a one-million-token context, with an input price of 0.9 yuan per million tokens and an output price of 3.6 yuan per million tokens. At the same event, iFLYTEK also
Synaptics has launched the Astra SR200 series of multimodal edge AI processors, integrating the Coral NPU architecture developed in collaboration with Google Research. The series includes three models, the SR210, SR230, and SR250, with AI compute of 200, 300, and 400 GOPS respectively, targeting AI glasses, smart wearables, smart cameras, and industrial vision. The products use a dual-processor architecture with Arm Cortex-M85 and Cortex-M52, integrate up to 8MB of SRAM and custom DRAM expandable to hundreds of MB, and support on-device execution of models such as Transformers. The two parties also launched the Coralboard SR development board equipped with the SR250 and the open-source Torq AI compiler, supporting on-device deployment of models such as Gemma 3 270M. The series has now been officially released, with no further updates at this time.
Synaptics has launched the Astra SR200 series of multimodal edge AI processors, integrating the Coral NPU architecture developed in collaboration with Google Research. The series includes three models, the SR210, SR230, and SR250, with AI compute of 200, 300, and 400 GOPS respectively, targeting AI glasses, smart wearables, smart cameras, and industrial vision. The products use a dual-processor architecture with Arm Cortex-M85 and Cortex-M52, integrate up to 8MB of SRAM and custom DRAM expandabl
On October 16, 2026, the Ministry of Commerce and the Ministry of Science and Technology issued an announcement soliciting public comments on the revision of the Catalog of Technologies Prohibited or Restricted from Export in China. According to The Decoder, the draft revision covers 23 key AI technologies including large models, agents, computing clusters and data processing; AI tools capable of autonomously writing or rewriting malware to attack critical infrastructure, military or government agencies are proposed to be fully prohibited from export, and some vulnerability-discovery AI tools face new export restrictions. The report states that large models are regulated when trained on and providing services based on China's important data, or when involving core technologies such as weights, checkpoints, source code, architecture and training and inference methods, with trigger thresholds of training compute reaching 10^26 FLOPs, at least 2 trillion parameters, or entering the global top ten; releasing restricted AI technologies and weights as open source can obtain automatic licensing, use through APIs, applications, websites, SaaS or lightweight on-device versions is not regulated, while deployment on local servers outside China requires applying for a standard export license. Other reports state that Joint Announcement No. 48 of the two departments proposes adding a total of 40 technologies in the three major fields of artificial intelligence, biomedicine and photovoltaics to the catalog, including 23 in artificial intelligence, and immediately implementing interim registration management before the catalog takes effect. The catalog is currently still in the public comment stage.
On October 16, 2026, the Ministry of Commerce and the Ministry of Science and Technology issued an announcement soliciting public comments on the revision of the Catalog of Technologies Prohibited or Restricted from Export in China. According to The Decoder, the draft revision covers 23 key AI technologies including large models, agents, computing clusters and data processing; AI tools capable of autonomously writing or rewriting malware to attack critical infrastructure, military or government