Tencent's open-weight Hy4 flash: 220B (12B active), rated 2.79/4 vs 2.81 for Claude Haiku 5.5 in Tencent's blind test, at ~1/18 the cost per task.
Weak at hard coding. Too big for my 8GB GPU.
Full breakdown: https://t.co/J8QtZhtQPv
NVIDIA's Nemotron for Science: an open 120B model (12B active) + chemistry dataset + recipes, commercial use OK.
Good news for open source. Bad news for my 8GB GPU.
What it does and how it compares: https://t.co/ldJ8MP3Hxs
Recruiters read hundreds of cover letters. The AI-written ones all sound the same.
Part 2 of my IT job-hunting series: clean up your online presence, email, LinkedIn photo and CV first.
https://t.co/9N8bM3iQzV
Do you let AI write yours?
Qwen4-Max is out. On Qwen's chart it beats Kimi K3.1 Pro on 14/16 tests; Opus 5.5 still leads 10/16.
Output price per 1M tokens: $4.50 vs Kimi $10 vs Opus $20.
My table + test notes: https://t.co/d0mGydCW7T
Brisbane's Queen Street Mall is already decorated for Christmas 🎄 Tip: the giant gem baubles have little windows, so look inside for reindeer 🦌 Photos and where to find everything:
https://t.co/agQCUwW4Sm
Benchmarked llama.cpp on my RTX 4060 Ti 8GB, Oct builds vs newest (post-v1.0):
1B model, prompt reading: +23–24%
1B model, writing: +3%
27B model split GPU/CPU: no change
Faster if it fits in VRAM. Otherwise RAM is the limit.
https://t.co/60c1O1lp9x
#llamacpp#LocalAI
Grand Plaza Browns Plains already has its 2026 Christmas decorations up—including the new Santa throne, centre court spiral tree, and giant food court baubles. Worth a visit before the December crowds hit.
Photo guide:https://t.co/ZH4sryDdEy
#Brisbane#LoganQLD#BrownsPlains
llama.cpp v1.0.0 is out 🦙
The big change: breaking API changes now only come with a new major version.
Also: MoE expert cache for small GPUs, FP8 KV cache, faster Windows loading, new web UI.
What it means on an 8 GB GPU:
https://t.co/zdUOJiGuKS
#llamacpp#LocalAI
Runway open-sourced Praxis-1: a 27B robot control model, Apache 2.0.
46.2% on RoboLab-120, the top open robot model (Runway's numbers).
Run it at home? Needs 80 GB, or ~23.5 GB in 4-bit on an RTX 50-series card. My 8 GB 4060 Ti: no 🙃
https://t.co/IqEPYfKIsH
#Runway#Robotics