AI
-
New Models
Tencent releases Hy4 flash, a low-cost multimodal MoE model with a 1M-token context, and opens its weights under a licence with limits for big platforms
Tencent launched Hy4 flash on November 16, 2026 and published its weights a day later on GitHub, Hugging Face and ModelScope. The mixture-of-experts model activates about 12B parameters per token, accepts text and images…
-
Research
MIT Han Lab trains an open 27B reasoning model to operate real robots using only reinforcement learning in simulation, and open-sources the PEAR training engine
A team led by Jiaming Tang, Zhuoyang Zhang, Junxian Guo and Song Han released PEAR on November 13, 2026, with arXiv paper 2611.18562. Starting from Qwen3.8-27B and using no demonstrations, simulated robot-use reinforceme…
-
Research
OpenAI reports that coding agents now complete 46% of two-to-four-hour research tasks without human help
On November 18, 2026 OpenAI reported from internal data that agent success on two-to-four-hour research tasks rose from 23% to 46% between March and October, as median researcher agent spend hit $2,400 a day and Safety a…
-
New Models
LlamaIndex releases OCRia-1.5 document-parsing models in 4B and 2B sizes for LlamaParse
On November 18, 2026 LlamaIndex released OCRia-1.5 for LlamaParse in a 4B Agentic version and a new 2B Cost Effective version, reporting higher ParseBench-2 scores for tables, charts, forms and bounding-box grounding.
-
Business
Anthropic prices its Nasdaq IPO at $83 a share, above the marketed range
On November 18, 2026 Anthropic priced its IPO at $83 per share, above the $70 to $78 range, selling 980 million Class A shares with Nasdaq trading under the symbol ANT expected from November 19; TechCrunch reports $81.3 …
-
Safety & Ethics
OpenAI details its frontier-model safeguards across research, internal and external deployment after the Hugging Face incident
On November 18, 2026 OpenAI described how it contains, monitors and gates frontier models at each stage from training to public release, including a pause of its most capable models' training and evaluation from Septembe…
-
Infrastructure & Chips
OpenAI says its models cut chip area on Jalapeño and AMD designs, and partners with Synopsys on GPT-Synopsys
On November 18, 2026 OpenAI published results on using its models for chip design, including estimated area cuts of about 13% and 10% in two Jalapeño blocks and gains it says AMD verified. OpenAI is also training GPT-Syn…
-
New Products
NVIDIA RTX Spark compact desktops from ASUS, Dell and MSI go on sale with up to 128GB of unified memory
On November 18, 2026 NVIDIA said RTX Spark compact desktops from ASUS, Dell and MSI are available, with a Grace CPU, a Blackwell RTX GPU and up to 128GB of unified memory for local AI agents and models. Acer, GIGABYTE, H…
-
Agents
Google Labs introduces Stitch Loop, a long-running product agent that opens pull requests from codebase and live metrics
On November 18, 2026 Google Labs introduced Stitch Loop, an experiment in which an agent watches a codebase and live product metrics, proposes fixes, writes code in a cloud sandbox and opens GitHub pull requests that a p…
-
New Products
Google brings Genie 4 to Project Genie for AI Ultra subscribers, with a new agent and community gallery
On November 18, 2026 Google rolled out a new Project Genie experience to eligible Google AI Ultra subscribers, giving them first-time access to the Genie 4 world model, upgrades to Genie 3, a Project Genie agent and a co…