OpenAI resets GPT-5-Codex users’ limits after an expansion slowdown
After a slowdown during additional GPU-capacity deployment, the team resets limits for all GPT-5-Codex users as compensation.
SEARCH
Search English reports
Browse published English reports with original sources and content provenance.
AI-assisted · reviewed
Alibaba Cloud releases Qwen3-Next
An experimental Alibaba release exploring novel routing strategies that sharply lower the active-parameter ratio per forward pass.
Mistral releases Magistral Small 1.2
Mistral releases Magistral Small 1.2, a maintenance update to its 24-billion-parameter Magistral Small reasoning model.
Mistral releases Magistral Medium 1.2
Mistral releases Magistral Medium 1.2, a closed model tuned for throughput and queue handling in high-concurrency enterprise API workloads.
OpenAI releases gpt-realtime
OpenAI brings the Realtime API to general availability and releases gpt-realtime for production voice agents.
Nous Research releases Hermes 4
The fourth-generation conversational model from community developers, instruction-tuned on a large open-source base.
Google releases Gemini 2.5 Flash Image (Nano Banana)
Google released Gemini 2.5 Flash Image, code-named Nano Banana, for image generation, editing, and multi-image fusion.
DeepSeek releases DeepSeek V3.1
An architecture tune of V3 that improves MoE expert load-balancing and the stability of very large-scale distributed training.
Z.ai releases 106B multimodal model GLM-4.5V
Z.ai releases GLM-4.5V, a 106-billion-parameter visual-language model with stronger image-text alignment and spatial-coordinate understanding.
OpenAI releases GPT-5
OpenAI's fifth-generation flagship, introducing a new architecture and setting a new bar across general-purpose tasks.
OpenAI releases GPT-OSS
OpenAI's first hundred-billion-parameter foundation model released to the open-source community.
Anthropic releases Claude Opus 4.1
Anthropic releases Claude Opus 4.1, an incremental closed-model update tuned for complex mathematics and long reasoning chains.
Z.ai releases lightweight MoE model GLM-4.5 Air
Z.ai releases GLM-4.5 Air, a lighter 106-billion-parameter MoE variant optimized to reduce redundancy and improve inference speed.
Z.ai releases 355B MoE model GLM-4.5
Z.ai releases GLM-4.5, a 355-billion-parameter MoE model with updated expert routing and improved long-form structure generation.
ByteDance Seed releases Seed LiveInterpret 2.0
ByteDance Seed releases the end-to-end speech-to-speech interpretation model Seed LiveInterpret 2.0 and says it is publicly available through Volcano Engine.
Alibaba Cloud releases 480B MoE coding model Qwen3-Coder
Alibaba Cloud releases Qwen3-Coder, a 480-billion-parameter MoE model for complex software engineering and long-form code generation.
Google releases stable Gemini 2.5 Flash-Lite
Google releases the stable Gemini 2.5 Flash-Lite model for high-throughput workloads that prioritize low cost and low latency.
OpenAI merges Operator and deep research into ChatGPT agent
ChatGPT agent unifies web operation, deep research, code execution, and editable deliverables into a single agent.
NVIDIA releases Canary-Qwen-2.5B
NVIDIA releases a 2.5B English speech-augmented language model combining FastConformer and Qwen.
Moonshot AI releases Kimi K2
Moonshot AI's trillion-parameter MoE model, continuing its emphasis on lossless recall across a very long context window.