@naaackers: I learned the expensive way that a monthly ChatGPT subscription and API usage are separate buckets. While experimenting with OpenClaw, I burned through my subscription allowance and then close to $40 in additional API tokens just getting the system set up. That token bleed is exactly why local models and running AI on your own machine can become appealing. local AI costs, OpenClaw API token usage, ChatGPT subscription versus API, self hosted AI

Naaackers | Tech Edu
Naaackers | Tech Edu
Open In TikTok:
Region: US
Monday 21 September 2026 18:00:41 GMT
4293
91
20
1

Music

Download

Comments

redgriffen18
Griffen :
How is it local AI if you are sending api tokens?
2026-09-23 03:39:30
0
aespa_yu_jimin
Alesana :
wild concept, don't use frontier overpriced American models and use DeepSeek flash.
2026-09-21 19:06:37
3
pajamahut
pj97 :
could just read the manual 😂😂😂
2026-09-22 04:27:30
0
de_struisvogel
struisvogel :
If you are using claud you are not using local AI. You don't burn "tokens" with local run AI
2026-09-21 20:39:35
2
cryomaner
cryomaner :
Depending on what you’re doing you don’t really need a large model either. 4b and 8b models can do a lot of menial tasks relatively accurately and can be ran on pretty economical hardware.
2026-09-22 21:56:43
1
horrorfreak1409
HorrorFreak1409 :
Openrouter has helped my affordability immensely! I still use Kiro to access the frontier models from OpenAI and anthropic but I use free models for agentic systems that can take hours, days, etc
2026-09-21 23:42:06
2
arsenicx2
Arsenicx2 :
Your not running local. Your just using chat GPT with extra steps 🤣😂🤣
2026-09-22 10:07:55
1
lelanddutcher
Leland :
you oauth you chat subscription to use that do not also add an api key and pay for tokens.
2026-09-21 19:45:17
1
robertheadley
Robert Headley :
Deep seek flash 4.1is my workhorse
2026-09-22 03:13:46
1
plebempire
pleb :
youre paying for ai broskie thats ur first problem
2026-09-21 19:44:11
1
dav_nettt
dav_nettt :
GLM 5.3 flash is crazy affordable on openrouter, and you can even get an $18/month sub directly from z.ai that definitely outputs more work than your Claude plan can for $20, and they are ok using it for OpenClaw and Hermes. best thing to do is deploy omniroute, and set up a combo that will use your z.ai sub first, the fall back to openrouter once your session limit is reached. also, check out Poolside Laguna S 2.1. it's not the smartest, but it does surprisingly well at long agentic workflows without loosing the plot, so it'll get there eventually. Openrouter has it as a free option with smaller context window and free tier limits that throw a 429 when you max the session.
2026-09-23 03:21:58
1
To see more videos from user @naaackers, please go to the Tikwm homepage.

Other Videos


About