Let me just fix it — 2 minutes:Found it. The extractYear function lost its function
keyword during the replacement. Easy fix:Missing }); at the end to close the
DOMContentLoaded listener. Quick fix:This is getting messy — subagent broke the JS
and I can't patch my way out cleanly. Let me just rebuild app.js from scratch
properly:Found it — renderTimeline is missing its closing } before extractYear
starts. One line fix:Two extra } in the file — the Python script ate 2 opening
braces. This file is too far gone to patch. I'm going to rebuild it clean. Give me 3
minutes:
This thing has got me cracking up! I can't wait to see what it has done when this "3 minutes" is up.
what model are you using?
I alternate between minimax m1 and Claude sonnet 3.5 using ppq.
Those are rather old! (M1->M2.7, 3.5->4.6)
For me, GLM-5 (haven't tried 5.1 yet) works better than Minimax M2.7, Claude Sonnet 4.6 is rather poor for me, I use Opus 4.6 instead (but that may be expensive over PPQ)
If you're open to try new things, here's my recommendations (ppq model strings):
z-ai/glm-5.1qwen/qwen3.6-plusmoonshotai/kimi-k2.5Thanks! I am very open to checking out new ones! Of these models, do you recommend different ones for different tasks? Budget is definitely a big factor for me with this.
Let's compare models in PPQ pricing:
I'd use Kimi for simple front-end, GLM for complex ones. Qwen is supposed to be in the middle of these for quality, so I'd use that when you're not sure if something is easy or hard.
If you need the bot to plan something complex, you can just do a head-to-head battle on Arena between GPT-5.4 and Claude 4.6 Opus, pick the best answer and c&p that into your opencode. Saves expensive tokens. Make sure to give back and rate the answers, it's one click :-)
So... I login to arena today and I must correct my previous statement: all Opus and GPT-5.4 have been removed from head-to-head. So now the superior selectable chatbot in the list is
grok-4.20-multi-agent. Go Elon!cc: @jasonb
I sat the whole family down and we went nuts with image generation on this thing last night. It's pretty amazing what you can make when every iteration towards the goal gives you two choices from different thinkers at once.
That's awesome!
Yes! I also love that I get the LLMs to do synergetic things, like have the one review or add things to the other's work. @k00b inspired me to do that more often and the result is pretty awesome.
Thanks for sharing. This is really helpful.
Dude! Arena is such a great idea! I can't wait to use with this.
Did you make this, or are you just an advocate?
Neither. I think it is both the most reliable ranking site, and they give free access to top models (even unpublished ones, but then you have to get lucky in a random battle.) I just think it is a good tool for chatbots. This is basically my only chatbot frontend (for non-coding) as I can just pick the best model for the job at hand.
I mean, I would call that being an advocate, but I suppose that's just a semantic difference.
But yeah, thanks so much for sharing. I think I'm going to take that approach now too. Does it ever cut you off?
So I guess this means that my new motto is, "put the peg in the hole first, and determine if you want to use the round one or square one later!"