ChatGPT GPT-6 Pro now splits a task across parallel agents
OpenAI's Oct 16 GPT-6 Pro refresh also speeds responses and tool use. Plus: an Apache-2.0 vision model that runs on a phone, and a free GLM playground.
Copy markdown
Fan one prompt out across agents, inside the chat tab
GPT-6 Pro can now break a task into subtasks that run in parallel, with each agent's progress and finished work openable right in the thread. It's rolling out to eligible Pro, Business, and Enterprise accounts (not Plus). Independent testing clocked a research brief at ~9% faster run-to-run (4m25s vs 4m51s) with no clear accuracy edge yet, and OpenAI hasn't documented an agent cap or whether a parallel run counts as one message or many against your limits.
MiniCPM-V 4.7 puts a real vision model on your phone
OpenBMB shipped the MiniCPM-V series' first mixture-of-experts build: 35B total, ~3B active, Apache-2.0. The team reports 13+ tokens/s in about 3GB on an iPhone 17 Pro Max for the quantized build. It does native detect, point, caption, and query with structured output, ships FP8 and GGUF quants, and serves day-0 on vLLM, SGLang, llama.cpp and Ollama — enough to build on-device document and camera features without a cloud bill.
Z.ai opens a free, no-key GLM playground
Z.ai put its MIT-licensed GLM Base, Reasoning, and Rumination models into a browser playground you can chat with for free — no API key, no local setup. It's a zero-friction way to size the models up before you wire them into anything.