Kimi and GLM Challenge Closed Source AI With Speed and Cost Advantages

Haha - this is unbelievable!

I suspect you just need Opus 4.8 and GPT 5.6 for just 10-15% of tasks

Kimi and GLM will do the rest

Enterprises who are paying eye-watering Anthropic bills are officially OVER PAYING like crazy 🤯

@bindureddy 1. Use high-thinking SOTA model (e.g. gpt-5.4 pro) for planning. Ask it research the topic and make at least 1000+ lines PLANS.md + AGENTS.md. 2. Use cheaper model e.g. Kimi K2.6, DeepSeek V4 or GLM-5.2 to actually write & verify most of the code. 3. Choose your harness well.

@KuittinenPetri lol! Choose abacus ai

@bindureddy Q I've built a hardware gate that makes LLMs provably incapable of emitting logical rule violations. A100 demo: 101B checks/sec, token gating in microseconds, zero failures. @bindu_reddy this could be a new trust layer for enterprise agents. 15‑min screencast if you want to see.

@bindureddy it’s gotten crazier that most of these top companies are now monitoring their token usage because of how expensive these tokens have become.

@weijianzhang_ or you need a very smart router :)

@bindureddy I agree with this. But it is still the "when to use which" question. I'm still biased to use the most powerful intelligence wherever possible...

@bindureddy GPT, KIMI and GLM is smashing it for me Looped Fusion running on #PIAgents

@bindureddy the reason most people still think closed source is viable is because they've never tried to scale something from scratch before. kimi's open model is the only way to truly accelerate innovation, not just get an early mover advantage

@bindureddy I think fable 5 is restricted for this. I will use it only orchestration and finding good loops.

@bindureddy As a non-company paying a plan though, it seems to me GPT is way cheaper

@bindureddy the overhead for big models really is getting absurd though

@bindureddy Open source is the only way forward for true decentralization and unstoppable tech. Cost and speed are just the beginning

@bindureddy Commoditize the 85%, innovate the 15%. Chinese models democratize AI; frontiers push boundaries. Hybrid wins

@bindureddy Abacus: write once, runs forever. Modern AI: deprecates next Tuesday.

@bindureddy True - OpenAI and Anthropic will be true research oriented intelligence companies in near future and oss models will serve for general usage

@bindureddy Kimi and GLM are growing most faster than anthropic wtf

@bindureddy enterprises were definitely paying the pioneer tax on inference costs

the real trend is heterogeneous inference becoming standard

@bindureddy I am subscribing and using Abacus AI, but I do use a LOT of other AI services as well. For coding I have tried Codex, GitHub Copilot, Cline, OpenCode, DwarfStar and Grok build. Sooner or later I will make my own AI coding agent in C++ (yes, it is going to be different).

@bindureddy Companies blindly chase top closed-source models while domestic LLMs handle most routine work fine.