📅 AI Industry News — November 13, 2025
Model Updates and Releases
OpenAI releases GPT-5.1 — Rolling out in ChatGPT, with API access available later this week. Includes 5.1 Instant (more conversational, improved instruction following) and 5.1 Thinking (adaptive reasoning). GPT-5.0 becomes a legacy model with sunset in 3 months. > OpenAI announcement
WeiboAI model released — A 1.5B parameter model based on Qwen2.5. Good initial performance, but drifts after 1–2 turns and can recite Quora content verbatim. > Related tweet
Gemini 3 Pro rumors — The community expects Gemini 3 Pro to launch next week; reported tests show performance close to human level, making GPT-5.1 look comparatively weaker.
Kimi K2 performance — Kimi K2 performs well on one-shot coding tasks, but API quotas deplete easily; a Cursor-like plan is recommended. > Demo video
---
AI Applications and Use Cases
Code Arena launches for coding evaluation — Evaluates models' coding abilities, supporting Claude, GPT-5, and more, requiring planning, debugging, and app deployment. > Twitter announcement
Perceptron launches Physical AI platform — Offers an API and Python SDK supporting Isaac-0.1 and Qwen3VL-235B for multimodal perception and action; free access to Isaac this week. > Perceptron announcement
Waymo expands autonomous driving coverage — Freeway driving in Phoenix, LA, and the SF Bay Area, connecting SF and San Jose with SJC curbside service. > Waymo announcement | Jeff Dean tweet
User wins custody case using ChatGPT — A user without a lawyer used ChatGPT to prepare legal documents and won full custody; the opposing party had an assault record. > Reddit post
Dawn newspaper uses ChatGPT to write articles — Pakistan's Dawn used ChatGPT for articles, raising concerns about AI-generated content; heavy editing distorted the output (30 em dashes). > Reddit post
UBTech releases self-charging humanoid robot — $112M in orders, with 500 units to be delivered by year-end for factory use. > SCMP article
---
Tools and Frameworks
LangChain introduces agent governance middleware — Human-in-the-loop (pause execution pending user approval) and tool call limits (preventing abuse). > Human-in-the-loop | Tool call limits
LlamaIndex releases SEC Agent template — A multi-step agent that classifies documents, extracts schemas, and audits SEC filings. > Twitter announcement
Cursor launches Max mode — Reads entire files; costs double beyond 200k context, and users suggest adding limits. > Discord discussion
Ollama simplifies local model running — Supports GGUF files for running models locally; link goes to Hugging Face docs. > Hugging Face documentation
Tokenflood released for LLM load testing — Open-source tool that simulates LLM workloads and evaluates latency. > GitHub repository
SMOLTRACE benchmark framework released — Evaluates Smolagents, tracking accuracy, latency, and more across 132 tasks. > GitHub repository
---
Community Discussion and Feedback
Mixed community feedback on GPT-5.1 — Some find it more conversational and helpful; others find it verbose and overly friendly. > Negative feedback | Positive feedback
Cursor vs Copilot user preference debate — Some users are returning to Copilot; preferences remain subjective. > Discord discussion
AI ethics discussion: privacy and propaganda — OpenAI fights NYT privacy invasion; AI chatbots may spread propaganda. > OpenAI blog | Discord discussion
Hardware discussion: CUDA vs Vulkan — Vulkan caused BSODs in LM Studio; switching to CUDA fixed it. Vulkan is faster but less stable. > Discord discussion
---
Other News
Windsurf releases Aether stealth models — Aether Alpha/Beta/Gamma, free to test. > Download link
Magic Patterns 2.0 raises $6M — Bootstrapped to $1M ARR, used by 1,500+ teams. > Twitter announcement
---
*Source: Easy AI educational project*