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
Gemini ❤️
claude
Yeah they're still tokens
That doesn't help if I forget the word itself. LLMs help turn a rambling stream of thought spewing out "it's kind of like X but not really" into "You must be thinking of the word...".
I've been leaning heavily on Claude 3.5 Sonnet and Gemini 1.5 Pro for the heavy lifting lately. Claude is still the king for refactoring messy deployment scripts, but I've been using Gemini 3 Flash via API for a few internal automation tasks where I need speed over deep reasoning.
I've also been playing around with DeepSeek V3 on a few local KVM slices to see how it handles specific Nginx edge cases. It’s surprisingly solid for the price-to-performance ratio.
@jar That log obfuscation system sounds like a solid use case. I've been thinking about something similar for automating initial ticket triage without giving an LLM the "keys to the kingdom."
Testing out Google Gemma 4 26B A4B model locally on my Macbook Pro M4 Pro 48GB ram and getting 50-56 tokens/second. Did a write up at https://ai.georgeliu.com/p/running-google-gemma-4-locally-with
Just started using Claude Cowork on MacOS and loving it so far. I updated my Claude Code AI image creator skill so it now works in Claude Cowork too. Can now ask Claude to generate images via Google Gemini 3.1 Image Flash (Google Nano Banana 2), Flux 2, Riverflow 2 Pro, Seedream 4.5, OpenAI GPT-5 Image https://ai.georgeliu.com/p/building-an-ai-image-creator-skill
Codex GPT 5.4 xhigh for cloud and OpenCode Google Gemma 4 for local
claude opus and gemini
As Claude's API user, I have to ask... Does $200/m max subscription worth the price tag today for web or cli use?
We need a LowEndCode forum aimed specifically at vibe coders and a BotVerify panel to benchmark the bots.
I think someone could earn some fast BilohBucks doing that.
Depends on the revenue earned or money saved from it.
It's definitely worth it even on the reduced peak-hour limit. If you use quite enough tokens to warrant it, then yeah, it's definitely worth it. You'll be able to use more than that $200 value. I'm on the Claude Max Plan, $100 and using between $4,000 and $6,000 a month in value.
Though the past week, instead of using Claude Code much, I've been using Claude Cowork and built a few interesting things.
I've been playing with Hermes Agent with self hosted Honcho as memory. Both using a MiniMax 2.7 API subscription. I haven't been very impressed yet in terms of MiniMax. I didn't expect it to be Opus 4.6, but it's not even Composer 1.5 (Kimi K 2.X?). I haven't worked with Hermes enough to see how well the memory/dreaming works. Have hit my Claude session limit three times today, which is a bummer.
Free ChatGPT is more than good enough for me at the moment. For questions where it's too shy to answer (such as many things related to sex), I also use DeepSeek, it appears much less censored.
Seriously? I haven't tried it yet.
Personally, I haven't tried it. Because even on API, there's performance differences during peak hour... But, judging from screenshot, it's bad enough for conversation and subpar performance for coding
https://cdn.sanity.io/files/4zrzovbb/website/037f06850df7fbe871e206dad004c3db5fd50340.pdfSee page 195-196
GPT
Ouch, that's pretty ugly.
EDIT: It looks like they are claiming that they are no longer optimizing for whatever that benchmark tests for. I can't actually test the thing because my Anthropic sub is out of token until Sunday and my API usage via cursor is done until the end of the month. I guess I could use the work subscription, but I can't do anything fun with that one.
Opus 4.7 looking good for me so far. Though you need to adjust the way you use it. Claude Code creator, Boris has some tips posted on Threads that I wrote about at https://ai.georgeliu.com/p/six-things-to-change-in-your-claude
I wrote a per session token usage/cost analyzer skill to inspect Claude Code sessions to see where consumption is and Opus 4.7 results at bottom of https://ai.georgeliu.com/p/i-built-a-token-cost-analyzer-skill
Insights into Claude Opus 4.5 vs Opus 4.6 vs Opus 4.7 vs Sonnet 4.6 testing effort levels from low, medium, high, xhigh, max https://ai.georgeliu.com/p/tested-claude-ai-llm-models-effort ^_^