OPENCODEX / Ice hockey / 24 Please call me JUN btw
Seoul
Joined February 2025
- Tweets1.7K
- Following439
- Followers1.3K
- Likes1.8K
trycua/cua is hands down the greatest open-source Computer Use project right now.
but there is one problem: it was built for cloud sandboxes and benchmarks, not for your personal laptop.
after forking T3 Code to build GOAT, we needed agents to actually use our own Macs and Windows PCs without breaking our workflow. so in the last 3 days, we pushed 68 PRs and 50,000+ lines of code into our cua-driver fork and runtime.
here is what we fixed to make it work on a real personal computer:
1. 10x faster screen reading: instead of asking the OS about each button one by one, it reads the whole window in one shot.
2. don't steal my mouse: it clicks and scrolls apps in the background across macOS Spaces while you keep working.
3. don't leak my secrets: passwords never touch your main clipboard, OTP fields are hidden from the model, and password managers (1Password, Proton Pass, NordPass) are blocked.
4. no more freezes: if an app hangs or has a 10,000-row table, the driver cuts it off in milliseconds instead of locking up.
5. no zombie processes: close the app and every background helper dies instantly.
6. real app permissions: on Windows and macOS, the agent can only touch the exact apps you approved.
cua gave open source the engine. we built the seatbelt and steering wheel for your actual machine.
GOAT is coming very soon.
I dont accept crypto
This guy didn’t redirect fees
He added use paid link
His real GitHub is github.com/lidge-jun/opencod…
And x acc of dev of 18k star is @claudeebum I’m in his DMs
only @KimiDevs chasing frontier!
Announcing Artificial Analysis Intelligence Index v5, featuring 2 new benchmarks and 3 upgrades. For the first time, the majority of the Index is built on private datasets
The Artificial Analysis Intelligence Index is our composite measure of model capabilities across tasks reflecting real-world usage today. v5 is our biggest upgrade since v4 launched in January. With v5, 60% of the Index is now based on evaluations with private test sets or held-out answers.
Alongside Intelligence Index v5, we’re launching the Artificial Analysis Progress Index v1, which tracks model capability over time on one continuous scale.
New to the Intelligence Index:
➤ FrontierCode 2.0: Cognition's newly-launched private benchmark of real-world coding tasks, graded on whether a repository maintainer would merge the agent's change
➤ Terminal-Bench-Science: 70 agentic scientific research tasks, run in a terminal environment
Upgraded:
➤ HLE-Diamond (replacing Humanity's Last Exam): a refined 1,000-question version of Humanity's Last Exam, split evenly between reasoning and knowledge questions
➤ AA-Omniscience v1.1: An updated dataset with saturated and public set questions removed, new questions added, and ambiguous or erroneous questions revised
➤ CritPt v2: a private, 40-question hard subset of CritPt v2, revised by the CritPt authors and run with a code-execution tool
Staying in the Index: AA-Briefcase, AutomationBench-AA, GDP.pdf and Terminal-Bench 4.0
Removed from the Index: GDPval-AA v2.1, SciCode, and AA-LCR v1.1
Key results:
➤ Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max, 57) and GPT-6.1 Sol (max, 55): Anthropic and OpenAI hold the top five models. Gemini 4 Argon (high) is next at 50
➤ The open weights gap widens from 3 points to 12: Kimi K3.1 Pro (weights not yet released) is the top open weights model at 46. Among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32 - 26 points behind the leader and below Claude Haiku 5.5 (max, 36)
➤ Cost per task varies by more than 13x among the leading models: of the 10 highest-scoring models shown, GPT-6.1 Terra (max) is the cheapest at $0.96 per Intelligence Index task and Claude Fable 5.1 (max) the most expensive at $13.04
Korean builders are some of the heaviest agent users on this app, and they're exhausted.
No real backing, no programs, just people burning their own money and weekends to keep building.
Then the first Grok Bot Meetup Seoul finally happens:
a stamp rally, an avatar booth, a quiz, and a macaron if you beat the mini game.
Same week elsewhere:
Paris: 60 builders
Toronto: builders night + dinner, 232 going
Bali: SpaceX's head of field engineering doing live Q&A
Ankara: people walked out with a month of credits
An OpenAI community meetup in Suwon got 243 signups for 60 seats. The demand is here.
@SpaceXSI Korea is not a macaron market.
adding computer use and browser use
in t3 code fork
want to make this clean
general agents delegates to linux, windows, mac
Opencodex
天天拿 Codex 敲代码，前端那审美是不是看得人头皮发麻？OpenCodex 算是把 Codex 给撬开了，非 OpenAI 的模型随时能切，压根不用蹲官方施舍支持。
三步就够：
1️⃣ 走 Kimi K3 的 OAuth 先把前端设计盘明白
2️⃣ 收工换 GPT sol5.6 接后端
3️⃣ 想摸情报再切 Grok 4.5 去 X 上搜一圈
一个壳子轮着换脑子使，确实舒服。
🔗 github.com/lidge-jun/opencod…
Kimi K3.1 Pro dropped yesterday. You don't need to update opencodex to use it.
opencodex pulls the model list live from your provider, so kimi/k3.1-pro and kimi/k3.1-pro-highspeed already show up in the dashboard as "new". Flip the toggle and they're live.
Then pick it from whatever you already use:
- Codex (CLI or app)
- Claude Code
- Kimi Code
Same model, any harness. No reinstall, no config edits, no waiting for a release.
If new models should switch on automatically, set the new model policy to "on".
Congrats!!
Proud to be ambassador for kimi!
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro
