No point of hiring a soldier without giving him weapons, or a professional driver without a car!
Surprisingly, This is where most of people lose it when dealing with their AI agents.
Most people setup their AI agents and then, wow it’s fun. But what should we actually do with it!
The real power of AI agents has n general isn’t only about their skills or architecture. It’s more about the tools they are connected to.
If you connect agents to your daily workflow tools, and then made your agents find the correlations between scattered tools, this where you feel the magic!
Honestly, I love openclaw because it’s the first, and will always have that place. But Google invented “attention” for LLM and Openai, had the ChatGPT. Openai made the first frontier but Claude now has the throne.. interesting era it is!
No much difference between OpenClaw and Hermes except one core thing, Reliability. I never had to restart gateway nightmare one time since I had it so far!
I don’t have to. Claude Code can 🤣 I asked it to do the job, download Hermes and replace memories and everything OpenClaw had, then I re-authenticated couple of stuff nd here I’m. 100% reliability, much better with Codex, and getting my personal assistant agent actually assisting me.
I’m just surprised with the advancement of technology Kimi and other Chinese models is doing in a very short time and in a very very cheap prices. I’m wondering how exactly the US models will keep competing with this. I’m so happy with this competition, and so every user should be!
Meet Kimi K2.6: Advancing Open-Source Coding
🔹Open-source SOTA on HLE w/ tools (54.0), SWE-Bench Pro (58.6), SWE-bench Multilingual (76.7), BrowseComp (83.2), Toolathlon (50.0), Charxiv w/ python(86.7), Math Vision w/ python (93.2)
What's new:
🔹Long-horizon coding - 4,000+ tool calls, over 12 hours of continuous execution, with generalization across languages (Rust, Go, Python) and tasks (frontend, devops, perf optimization).
🔹Motion-rich frontend - Videos in hero sections, WebGL shaders, GSAP + Framer Motion, Three.js 3D.
🔹Agent Swarms, elevated - 300 parallel sub-agents × 4,000 steps per run (up from K2.5's 100 / 1,500). One prompt, 100+ files.
🔹Proactive Agents - K2.6 model powers OpenClaw, Hermes Agent, etc for 24/7 autonomous ops.
🔹Claw Groups (research preview) - bring your own agents, command your friends', bots & humans in the loop.
-
K2.6 is now live on https://t.co/YutVbwktG0 in chat mode and agent mode.
For production-grade coding, pair K2.6 with Kimi Code: https://t.co/uvoSJKyGCY
-
🔗 API: https://t.co/EOZkbOwCN4
🔗 Tech blog: https://t.co/9wWvgIQSS3
🔗 Weights & code: https://t.co/Be0hjs2RTP
Opus 4.6 “got dumber” for 3 main reasons:
1.Adaptive thinking sometimes allocates zero reasoning on hard turns
2.Default effort was silently lowered to 85 (medium)
3.Outages + instability made it worse
Fix: explicitly force deep thinking. In Claude Code, set effort to `high`/`max` (or use `ULTRATHINK`), and disable adaptive thinking via config so every turn gets a real reasoning budget.
We solved character consistency. Forever
Avatar V captures you in 15 seconds and holds your identity across every video.
Change the look, outfit, and setting to create unlimited versions of you.
RT + comment "AvatarV" below and I'll DM 100 credits to test it out (must follow)
@karpathy Honestly this makes much more sense than hidden memory. If I can see it and edit it then I can trust it. In real business use the problem is not memory only, the problem is wrong memory, once it remember wrong thing you spend more time correcting than working.