New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
Great to see more 2 million context token AI LLM models released! I tested both Sonoma Dusk Alpha, Sonoma Sky Alpha stealth models, along with Qwen3 Max and 10 other models for code analysis and ranked them all at https://github.com/centminmod/sonoma-dusk-sky-alpha-evaluation ^_^
Sonoma models do look pretty "ehh" across the board
Still Claude Code #1 for me.
Been dabbling with Claude Code OpenTelemetry https://docs.claude.com/en/docs/claude-code/monitoring-usage and Grafana dashboards. Screenshots of my Claude Code usage metrics collected in Grafana at https://github.com/centminmod/claude-code-opentelemetry-setup, including a Claude Code usage metrics MCP server so Claude can query its own usage metrics
A good mix of ChatGPT and DediRock AI™
Recently, I used “GPT-5 thinking” on Perplexity.
Normally, I use GPT-5 for searching and coding, Claude sonnet-4.5 thinking for self-discussions, GPT-4o mini for translation assistance, and Gemini 2.5 for the built-in model of my tools. Each model is running on several platforms(like OpenRouter, perplexity, the model’s official platform) For now, I can use them for free.
i guess you are from cn
Kimi-K2 selfhosted as the newish beast to try.
Normal gear: gpt-oss:120b and deepseek-R1-0528:671b.
qwen3-coder:480b is also on the server, would need some more tests
Claude, Grok, Gemini, and Copilot for small things that dont require a large context window.
Claude is my favorite though and daily driver.
Reminder that the 12-month free Perplexity Pro AI plan account when paid with PayPal or Venmo ends in less than 24hrs on December 31, 2025 https://www.ozbargain.com.au/node/922666.
Use link at https://www.perplexity.ai/join/p/paypal-subscription for PayPal or Venmo https://www.perplexity.ai/join/p/venmo-subscription
Also Kilo Code is looking to sponsor some open source projects if they use Kilo Code for AI code reviews https://form.typeform.com/to/KeAxDfj5
Coding:
(recently)
IDE: Google Antigravity with Claude Opus 4.5
Chatbots: https://gemini.google.com and https://claude.ai/
Other stuff:
(Fast and non-important): https://copilot.microsoft.com/
https://chatgpt.com/
https://chat.deepseek.com/
Claude Opus 4.5 & Gemini 3.0 pro
Claude Code still.
Video demo of my Claude Code hook for using Kokori 82M Text-To-Speech (TTS) model https://github.com/remsky/Kokoro-FastAPI to give Claude Opus 4.5 a voice in my AI coding sessions and notify me when it needs my attention. Been using this setup for almost a year now with Claude Code https://www.threads.com/@george_sl_liu/post/DTEj-HfEqzs?xmt=AQF0NbqD8yZeD2YN4cgW8Gp7i31iKdZnTgLo9So-FV8OCw
antigravity with opus for coding
veo3.1 for video gen
nano banana pro for images
everything else chatgpt.
This is cool! I didn't look at your repo but does this cause additional context usage in your sessions?
I imagine it might but not 100% sure. But it's only a few extra lines of text that is passed on as hooks only call the script. Not notice much difference in my Claude Code OpenTelemetry Grafana usage metrics https://github.com/centminmod/claude-code-opentelemetry-setup
Gemini & ChatGPT
gemini 3 pro
Gemini 3 Flash
I’ve found that while DeepSeek’s reasoning is interesting, the Claude + GPT combo is still the ultimate power user setup for work.
Gemini is great for the API and large context, but for actual production work, Claude has a much better human nuance and GPT is still more reliable for complex, multi step logic. DeepSeek feels a bit like a novelty in comparison, it’s cool to see the "thoughts," but Claude and GPT just get the job done faster without the fluff.
I am fisting my 64gig KS-LE-B mainly with 80B Qwen.
Have to setup GLM AIR again, was pretty good.
ChatGPT
DediRock AI™
gemini / openai
Also signed up with Z.AI Pro plan with 50% discount + 10% invite discount so $360/yr plan is now $129.60/yr https://www.threads.com/@george_sl_liu/post/DTSfoMLk092?xmt=AQF0ZAALqDMVDdruMFs1trmj40CXs7nLIoWivb0-o7bxiA invite code in my Threads' post ^_^
Updated my Claude Code starter guide with Z.AI pairing configuration setup info at https://github.com/centminmod/my-claude-code-setup#using-zai-with-claude-code
Is that a recurring discount or just for one billing cycle?
1st billing cycle only unfortunately. But still a good deal for next 12 months and i cancelled the auto renew billing so it ends in 12 months
One billing cycle.
I might want some AI help with building HTML and CSS landing pages. I'll have to see if Z.AI can help with that. Is there a better way?
Give it a try Claude Code + ZAI though for that probably Claude $20/month plan can do the job if difference in price isn't an issue
I am using Google Antigravity for that.