yemre
@agenticn3rd
Istanbul-based developer building practical AI agents, local-first tools, and industrial systems.
2.7K Following 159 Followers
GLM 5.2 is 5x cheaper than Opus 4.8 and 11x than Fable 5, yet it tops PostTrainBench.
That’s exciting because lower costs make personalized intelligence economically viable. Every company and country should be able to own models trained on its own data and have sovereignty over it. The future is millions of models, each crafted around the data, values, and decisions of the people who rely on them.
Show more
agentic coding when lid is closed. nice.
dude I love cafes. the coffee taste mid but a computer nerd like me has to touch grass... or tables
I want deepseek v4 flash on my thinkpad.
I dont know how tf it will fit but ee gotta make it happen
Man the feeling of feeling your feelings but u also don’t know what you feel.
Anyways did u know neural networks are fckn awesome?
I’m Yunus.
I build small AI tools because I like things that are useful, private, understandable, and a bit unusual.
is an on-device chatbot experiment without LLM APIs.
is a free browser-side toolbox with 100+ tools for AI, math, dev, security, SEO, and odd practical problems.
I care about computers, but mostly I care about making life less frustrating.
GitHub:
Show more
I’m Yunus.
I build small AI tools because I like things that are useful, private, understandable, and a bit unusual.
is an on-device chatbot experiment without LLM APIs.
is a free browser-side toolbox with 100+ tools for AI, math, dev, security, SEO, and odd practical problems.
I care about computers, but mostly I care about making life less frustrating.
GitHub:
Show more
"Founder is ex Google TPU team"
Introducing LongCat-2.0 🐱
1.6T parameters · MoE with ~48B active · 1M context
The full model behind Owl Alpha on @OpenRouter — now available.
Built for agentic coding from the ground up:
◆ LongCat Sparse Attention (LSA) — scales efficiently for 1M-context tokens
◆ Zero-Compute Experts — dynamic activation 33B–56B per token, zero wasted compute
◆ MOPD — three specialized expert groups (Agent / Reasoning / Interaction), gate-routed per task
How it stacks up:
→ Terminal-Bench 2.1: 70.8
→ SWE-bench Pro: 59.5 (GPT-5.5: 58.6)
→ SWE-bench Multilingual: 77.3
→ FORTE: 73.2 · RWSearch: 78.8 · BrowseComp: 79.9
📖 Tech Blog:
Try it across different scenarios 🧵👇
Show more
In 1950, Turing said: stop arguing about whether machines think, judge what they can do.
That is still the cleanest filter for AI:
less mysticism more output.
Jensen crying in the style of the native american . One single tear
the pain and suffering of mine..... i prompt and prompt and repeat and prompt and write prompts for prompts....
Lol @interaction your AI is a real beast , love it!
Clearly GLM 5.2 is much better at AI engineering and RL than my amateur self lol
2300
we’re still waiting for AI bubble burst while Dario warns us about Mythos 81 killing two of its researchers
tokenmaxxxing at its finest
a new era of custom harnesses begins....
I will make it completely open-source and free forever.
Galaxy Merge Harness, your Linux-native Web GUI harness.
Built-in tools, auto-skill discovery, multi-llm fusion system built-in and more.
This time (again) it will work (again). :)
Show more
