@teortaxesTex In my opinion, the "usage/(know-the-existence) of harness softwares" makes the biggest difference.
English
ChiL
980 posts
ChiL
@cdnewl
Coder, Game dev. Life isn't about how many punches you take, it's about how many times you take punches and move on.
U.S. Tham gia Aralık 2011
469 Đang theo dõi156 Người theo dõi
With this generation of agents, the gap in viscerally understanding AI capability between free tier chatbot users and power users is greater today than in the days of 3.5-turbo vs 4T
we live in different universes
English
@raju_datla1 @quxiaoyin GLM 5.1 I can understand, but MiniMax? I don't think MiniMax is anywhere close to DeepSeek v4 or GLM 5.1 or Kimi 2.6 tier TBH.
English
Yeah, probably fast and crappy. Other than Minimax and GLM 5.1, none of other chineese models really do anything consistently or good enough. But then, I guess if its working for you great. Unless you are just using for non-coding work or non-serious work, then yeah, probably less cost than claude
English
I can’t believe I stopped using Claude Code max and entirely use DeepSeek and Hermes. It’s so fast, so so fast, 3x faster for the same task. So cheap. I spent $5 last week and never need worry about being rate limited or usage hit limits very two hours. For most tasks it’s perfect enough.
English
@comingconflict @HealthRanger So, subsidizing the corporation with taxpayer money and NOT pass on the benefits to customers... is what you are suggesting...
English
I just had DeepSeek V4 find and fix 8 memory leaks that were causing crashes in code written by Claude Opus 4.7.
DeepSeek found them all and fixed them in minutes, at a total cost of maybe three pennies.
DeepSeek is amazingly good at coding and bug fixing. And it costs almost nothing to use, even the Pro version.
Best harness? OpenCode.
English
@comingconflict @HealthRanger Bruh, we have Stargate project, 1 Trillion USD, remember? That's like what, last year?
English
@HealthRanger If the US subsidizes domestic AI it would be cheap also. That’s the only reason deepSeek is so cheap
English
Follow-up on non-English token-inefficiency with more model-language pairs:
- Chinese is cheaper than English on major Chinese models
- Gemini and Qwen provide least non-English tax
- Anthropic has the highest tax by far; Kimi is next
- Hindi is the worst-covered language here, despite its massive speaker base
Aran Komatsuzaki@arankomatsuzaki
The non-English tax is real. Sutton's Bitter Lesson, translated across languages and normalized to OpenAI English token count: Hindi: OpenAI 1.37×, Anthropic 3.24× Arabic: OpenAI 1.31×, Anthropic 2.86× Chinese: OpenAI 1.15×, Anthropic 1.71× Claude’s tokenizer charges a much higher linguistic tax.
English
@teortaxesTex Normally I'd still prefer a coding/token plan. However, if the token price is low enough, I guess it makes sense just use it as is.
I guess there is one area pay-as-you-go is better: you get to use the API anywhere (your own app etc.) whereas subs usually limit usecase by ToS.
English
I've been saying that DeepSeek should do a subscription to grab some cash
but no, I guess their approach is better.
"We sell API tokens. Very cheap tokens, very many. If we want to do PR, we sell them cheaper. That's our way. No plans, no credits. Deal with it"
English
Hermes Agent tip of the day:
You can backup and transfer your agent cleanly and simply. Want to move it to a new VPS that is bigger? Take it to your new Mac Mini?
`hermes backup` - > creates a zip
install hermes on a new machine fresh
transfer the zip to that machine
then run `hermes import`
done :)
Full docs:
#hermes-backup" target="_blank" rel="nofollow noopener">hermes-agent.nousresearch.com/docs/reference…
English
@teortaxesTex This is perfect for my use case for real, as I used Hermes and OpenCode a lot, both of which benefits from cache hit massively
English
Cache hit rates + price slashes are so important
Agents don't just constantly reread their own outputs. The context grows *way faster* than that. This is 6.2M cache hits, $0,0224, literally 2 cents. Yesterday, it'd be $0,22. >50% of the total bill.
English
@teortaxesTex @Teknium Im currently using Hermes and OpenCode concurrently. I think they actually complement each other. "Right tool for right task" kind of thing
English
tried and got unironically impressed by Hermes Agent. They were turning into something of a merchandise/media company, but I guess the same instincts contribute to tasteful polished software. I like it more than pi now (nevermind the claw family). Congrats @Teknium et al
English
Introducing "create-agent-tui"
A skill for building your own agent harness + terminal UI (TUI). The skill walks you through 4 different ways of customizing the look, and supports dozens of optional features 👇
OpenRouter@OpenRouter
Stay tuned for an exciting openrouter/agent SDK tutorial, coming tomorrow! Start building on the top floor, on battle-tested abstractions that scale with model intelligence: openrouter.ai/sdk
English
GLM 5 -> 5.1 - Great improvement
Kimi K2.5 -> K2.6 - Great improvement
DeepSeek V3 -> V4 - Great improvement
Qwen 3.5 -> 3.6 - Great improvement
I am looking at you @MiniMax_AI. When 3.0? 😀
English
FINALLLY FINALLY it is here.
V4-flash: all the way back to V2 prices, only now with 1M
V4-pro: roughly Kimi/GLM/MiMo competitor
Chat prefix completion and FIM back – thank you! Missed this forever
but what can they do?
Lisan al Gaib@scaling01
DEEPSEEK-V4 FLASH AND PRO ITS HAPPENING api-docs.deepseek.com
English
you guessed it
(If you guessed "it has never went anywhere": that too)
China has an absurd, ever growing amount of research in materials and nanotechnology, on its face that should have amounted to some sci-fi products by now, and yet
Kirito (e/acc) 🏴☠️@bronzeagepapi
An in-memory computing architecture based on a duplex two-dimensional material structure for in situ machine learning nature.com/articles/s4156…
English
Iran just attacked three ships in the Strait of Hormuz
This is bullish for the stock market because these ships need to be repaired, causing steel and glass demand to go up resulting in GDP growth
English
