OpenAI announced it is strengthening ChatGPT's response to violence and self-harm risks along with its enforcement system.
DeepSeek cut V4-Pro pricing by 75% and also lowered API cache-hit costs to one-tenth of the previous rate.
DeepMind alum David Silver's Ineffable Intelligence has raised a record-breaking $1.1 billion seed investment.
Xiaomi has unveiled MiMo-V2.5-Pro, featuring enhanced agentic capabilities and long-horizon task performance.
NVIDIA B200 GPU spot prices surged 114% in six weeks.
The proliferation of open-weight models is shaking the monopoly-moat logic of US AI.
MIT researchers have proposed Recursive Language Models, which split up context to explore it.
ESRRSim found that the detection rate of strategic reasoning risks across 11 reasoning LLMs varied widely, from 14.45% to 72.72%.
OpenAI's GPT-5.5 raised performance, but its safety evaluation remains shallow.
For a single agent, the Batch API's latency is fatal, and cost efficiency only grows meaningful when multiple agents are bundled together.
This analyzes why AI application companies should adopt post-training based on open-weight models to cut costs and optimize performance.
China's NDRC has blocked Meta's acquisition of **Manus** and ordered the deal to be withdrawn.
Reports suggest OpenAI is developing a smartphone processor, with MediaTek and Qualcomm collaboration and mass production expected in 2028.
OpenAI missed its new user and revenue targets, adding pressure to its IPO preparations.
This work showed on CLEVR and SDXL that compositional generalization in Conditional Diffusion corresponds exactly to local conditional scores.
StereoFoley generated stereo audio that reflects object positions from video.
OpenAI and AWS have launched a limited preview of OpenAI models, Codex, and Managed Agents on AWS and Bedrock.
Microsoft's VibeVoice transcribed 1 hour of audio in 8 minutes 45 seconds.
This introduces a technique for building AI agent teams that review like actual Korean office workers by leveraging NVIDIA's 1 million Korean personas.
Across 36 test runs, v4-flash had zero downtime while v4-pro was the slowest.
The Musk-Altman trial has kicked off the full-scale battle over OpenAI's for-profit conversion.
Claude Pro users must pay for additional usage to use Opus in Claude Code.
Released Talkie, a 13B vintage LM trained on documents from before 1930.
On the Mahoraga benchmark, Qwen3 4B outperformed cloud agents.