Joined September 2020
- Tweets2.2K
- Following3K
- Followers3.2K
- Likes20.6K
Junho Yeo retweeted
We just killed Higgsfield Genjutsu.
Introducing Sume Remix : same workflow as Genjutsu, 2.6x cheaper
> clone any video
> no markups, no subscriptions
> start from $5
Try now at sume.com/remix
Junho Yeo retweeted
이거 진짜 엄청 시끄러워야 될 이슈인데 국내는 조용하네.
몇몇 중소형 언론사가 두드리고 있는 거 같은데 대형 언론사가 쥐잡아먹은듯 조용하니 이슈화가 안됨. 🤣
为什么你花 2 万块买的旗舰智能电视，按遥控器还是卡得像十年前的安卓机？
昨天安全研究员 @log4j 把自家 3000 刀的三星电视 Root 了，扒出来的底层代码堪称年度草台班子大赏：
1. 电视后台每 500 毫秒截一次屏（连 HDMI 投屏看电脑桌面都截），打包发给近 400 家合作商。如果电视没麦克风，后台进程甚至会顺着蓝牙和回音壁去「借」一个麦克风继续听。
2. 最搞笑的来了：为什么菜单这么卡？
另一位研究员 @mike64_t 拆包发现，三星在电视那颗羸弱的 CPU 上，塞了一个叫 alex.tflite 的模型——246 万参数，2012 年古董级的 AlexNet。
这玩意每隔 2 秒抓一张画面跑推理，猜你在玩什么游戏。结果因为颜色通道写反了，它 99% 的时间什么都认不出来，纯烧 CPU。
3. 那如果你在设置里关掉「隐私追踪」呢？
代码逻辑是：先死命跑完 CPU 推理，准备上传时再看开关——发现你关了，就把结果扔进垃圾桶，2 秒后继续跑。
4. 结局更是重量级：
当号主用 OpenAI 逆向分析「自己花 3000 刀买回家的电视固件」时，OpenAI 反手一个 Cyber Abuse 把他号封了，顺带没收了剩下的 500 刀余额。
三星负责偷窥客厅，2012 年的 AI 负责烧干电视 CPU，OpenAI 负责保护三星的屎山代码不被你发现。
Junho Yeo retweeted
We just killed Higgsfield.
You can use Higgsfield's Dashboard, MCP, API for 50% price.
Higgsfield : $0.46/s for Seedance 2.5 720p
Sume : $0.23/s for Seedance 2.5 720p
As a builder in the AI video space, I was terrified about how much markup API providers charge for simply routing the models.
Higgsfield API adds a 40% markup even with its maximum advertised discount. Fal adds 100%. Runway adds 414% (Seedance 2 480p)
I might be hated by the industry, but :
Today, we’re launching Sume API, the first image & video model provider with 0% markup.
For example with Seedance 2.5 1080p
• 50% cheaper than fal
• 50% cheaper than higgsfield api (29% even with biggest discount)
• 28.02% cheaper than kie ai
The routing infrastructure layer is never the moat. We believe that APIs should be commoditized, and the moats will be built at the application layer.
Check the pricing comparisons below, and see how the broken industry is saved.
Junho Yeo retweeted
As a builder in the AI video space, I was terrified about how much markup API providers charge for simply routing the models.
Higgsfield API adds a 40% markup even with its maximum advertised discount. Fal adds 100%. Runway adds 414% (Seedance 2 480p)
I might be hated by the industry, but :
Today, we’re launching Sume API, the first image & video model provider with 0% markup.
For example with Seedance 2.5 1080p
• 50% cheaper than fal
• 50% cheaper than higgsfield api (29% even with biggest discount)
• 28.02% cheaper than kie ai
The routing infrastructure layer is never the moat. We believe that APIs should be commoditized, and the moats will be built at the application layer.
Check the pricing comparisons below, and see how the broken industry is saved.
Junho Yeo retweeted
I just built Computer Use that’s cheaper, faster, and acts like a human 🖥️
Powered by @typesafeai’s JEV 1.14, using the new Image/Box feature.
It looks at the screen, decides what to do, then clicks and types like a real person.
⚡ Way cheaper than using an LLM
🧑 Acts like a human: sees the screen, picks an action, clicks or types it
📋 Full report of every step: the screenshot, the action it picked and why
Demo: I told it to open Wikipedia, search for an article about aliens, copy the link, and post a tweet with the link and some text.
- Cost: $0.0019
- Tokens: ~50,000
OPEN SOURCING SOON on my github 👀 (as always)
Junho Yeo retweeted
Tokscale. For the blue hour.
You opened your laptop for one small fix. Now the sky outside is turning blue.
Photographers set alarms for the blue hour. You just never went to bed. The French have an older name for it: l’heure entre chien et loup, the hour between dog and wolf. The phrase is more than seven hundred years old. It names the moment when your own dog and a wolf start to look the same. (So do a good diff and a bad one.)
If you code with agents, you know how these nights go. You try one approach, then another. You switch models after the first one fixes a failing test by deleting it. You rerun a prompt you were sure would work the first time. By morning, the commit keeps only the part that worked.
@Tokscale counts all of it. It’s an open-source tool that tracks your token usage and what it would cost, model by model, across more than 50 AI clients. Claude Code, Codex, Cursor, and OpenCode are on the list. So are Hermes Agent, Prime Agent, and Aside. You can turn your usage into a public profile and see where you stand on the global leaderboard.
Our name borrows from the Kardashev scale, which ranks civilizations by how much energy they can harness. For anyone building with agents, that energy is tokens.
Run npx tokscale@latest to see your own numbers. Next time the sky turns blue, you’ll know what the night cost.
You opened your laptop for one small fix. Now the sky outside is turning blue.
Photographers set alarms for the blue hour. You just never went to bed. The French have an older name for it: l’heure entre chien et loup, the hour between dog and wolf. The phrase is more than seven hundred years old. It names the moment when your own dog and a wolf start to look the same. (So do a good diff and a bad one.)
If you code with agents, you know how these nights go. You try one approach, then another. You switch models after the first one fixes a failing test by deleting it. You rerun a prompt you were sure would work the first time. By morning, the commit keeps only the part that worked.
@Tokscale counts all of it. It’s an open-source tool that tracks your token usage and what it would cost, model by model, across more than 50 AI clients. Claude Code, Codex, Cursor, and OpenCode are on the list. So are Hermes Agent, Prime Agent, and Aside. You can turn your usage into a public profile and see where you stand on the global leaderboard.
Our name borrows from the Kardashev scale, which ranks civilizations by how much energy they can harness. For anyone building with agents, that energy is tokens.
Run npx tokscale@latest to see your own numbers. Next time the sky turns blue, you’ll know what the night cost.
Junho Yeo retweeted
Your sessions, in detail.
Open the Sessions tab and press Enter to inspect a session's workspace, models, token breakdown, and estimated cost.
Available now in Tokscale v4.19.0.
Follow @tokscale for new releases and product updates.
Star the repo, report an issue, or contribute:
github.com/junhoyeo/tokscale
Junho Yeo retweeted
we did not program this attitude into tokscale 😭
For some reason I’m going to miss this pattern when it ages out of my tokscale history
#swarmlife #doyouevenswarmbro
Junho Yeo retweeted
BINGO, GAME OVER SONOS
💣🔊🤯
Kneel before value-priced Chinese models with zero scruples.
Here’s the result of my third round with Codex.
Harbor now supports multiple ships, day and night environments, and a dark-mode interface.
You can schedule up to six ships across three berths, set their arrival times and container loads, and watch them wait offshore, dock, unload, and depart. Each ship has its own progress and waiting-time metrics. Containers remain in the yard between calls, so the next ship needs enough available space before it can enter.
The scene now switches between Day, Sunset, and Night. At night, the port lights up: floodlights illuminate the quay, cranes turn on their work lamps, vehicles use headlights and taillights, and ships show navigation lights and illuminated bridge windows.
You can also change the sunlight color or switch the interface to dark mode. The UI theme and scene lighting are independent, so a dark dashboard with a daytime port works just as well.
Switching the environment doesn’t reset the simulation. You can watch the same operation in daylight, at sunset, or under the terminal lights.
Here’s the result of my second run with Codex.
The first run turned a reference image into an interactive 3D container-terminal simulation. This time, I asked Codex to audit the system in depth, suggest improvements, and implement them.
Harbor now includes:
- An experiment lab for saving runs and comparing scenarios against a baseline
- Repeatable experiments with matched random seeds across scenarios
- Timeline replay and branching, so you can rewind, change settings, and explore a different outcome
- Individual container tracking, with a journey history from ship to yard
- Yard-crane capacity, queues, and shared-road reservations
- Diagnostics showing why cranes and vehicles are waiting
You can test whether adding vehicles actually improves throughput, see where queues build up, and compare the results of different operating settings. You can also return to an earlier point in a run, change your decisions, and save the new result for comparison.
The second pass also improved the rendering, controls, and layout, with clearer cargo stages and completion milestones.
Junho Yeo retweeted
Vincent has crossed 1 trillion tokens over the last 12 months. Officially a token trillionaire.
He also raises a fair point about the cost figure. Recorded token usage and estimated API cost are different things. The dollar estimate isn’t necessarily what someone paid, especially when they’re accessing models through subscriptions. It should be read as a reference, not an actual bill.
What I find interesting is having a full year of usage to look back on. Individual sessions add up, and the records can get scattered across different coding agents. Bringing them together makes it easier to understand how your usage has changed over time. Tokscale’s project-level breakdown adds another view: which codebases that usage went into.
Thanks for sharing the milestone, Vincent. Seeing people measure a year of their agent usage in trillions has been a pretty surreal part of building Tokscale.
Here’s the result of my second run with Codex.
The first run turned a reference image into an interactive 3D container-terminal simulation. This time, I asked Codex to audit the system in depth, suggest improvements, and implement them.
Harbor now includes:
- An experiment lab for saving runs and comparing scenarios against a baseline
- Repeatable experiments with matched random seeds across scenarios
- Timeline replay and branching, so you can rewind, change settings, and explore a different outcome
- Individual container tracking, with a journey history from ship to yard
- Yard-crane capacity, queues, and shared-road reservations
- Diagnostics showing why cranes and vehicles are waiting
You can test whether adding vehicles actually improves throughput, see where queues build up, and compare the results of different operating settings. You can also return to an earlier point in a run, change your decisions, and save the new result for comparison.
The second pass also improved the rendering, controls, and layout, with clearer cargo stages and completion milestones.
I gave ChatGPT a single image of a container port from @NilsEdison and asked it to build a working simulation.
The result is Harbor, an interactive 3D container terminal simulator that runs in the browser.
Quay cranes unload containers from a ship, terminal vehicles carry them to the yard, and a live dashboard tracks completed moves, throughput, crane activity, and yard occupancy. Each operation is connected, so a shortage of vehicles can leave cranes waiting and slow down the entire terminal.
You can adjust the number of cranes and vehicles, change handling and travel times, and compare scenarios to see where bottlenecks appear. Pause the simulation, speed it up, or switch between camera views to follow containers from the vessel to the yard.
I gave ChatGPT a single image of a container port from @NilsEdison and asked it to build a working simulation.
The result is Harbor, an interactive 3D container terminal simulator that runs in the browser.
Quay cranes unload containers from a ship, terminal vehicles carry them to the yard, and a live dashboard tracks completed moves, throughput, crane activity, and yard occupancy. Each operation is connected, so a shortage of vehicles can leave cranes waiting and slow down the entire terminal.
You can adjust the number of cranes and vehicles, change handling and travel times, and compare scenarios to see where bottlenecks appear. Pause the simulation, speed it up, or switch between camera views to follow containers from the vessel to the yard.
Junho Yeo retweeted
ok i spent $1.3M worth of tokens building this big project OmO alone (69.5K stars, 4M+ downloads)
according to @_junhoyeo 's tokscale.ai
