Today's thread
What ties today's news together is a reason to look again at money you already pay and data you have already handed over. OpenAI added API credits to ChatGPT Pro and Business subscriptions and raised Codex usage, so the same subscription fee now covers more. Google's new Gemini 4 Carbon model is still off-limits to most people, but its price list is already out.
On the other side, Daishin Securities, Mirae Asset Global Investments and Bithumb each confirmed outside access to their systems. It is a day to check where your data sits and what kinds of messages it could invite. On the tools side, IeumOffice edits Hangul documents in their original format, and the new MacBook Pro takes up to 128GB of memory, widening the options for running AI on your own computer.
ChatGPT Pro gets up to $500/month in API credits
Starting October 28, OpenAI began giving monthly API credits to paying ChatGPT Pro and Business subscribers. The API is the doorway through which your own programs and automations call OpenAI models instead of the ChatGPT screen, and it used to be billed separately from the subscription. Pro gets $100, $200 or $500 a month depending on tier; Business gets $20 per Standard seat and $100 per Premium seat, pooled up to $500 a month per workspace.
To qualify, the paid subscription must be at least seven days old; free trials and complimentary subscriptions are excluded. On Pro, find the benefit under Billing in ChatGPT settings and choose an API organization (the account unit that API charges are billed to) that you own under the same email. The first grant is prorated for the rest of the billing period, and unused credits expire at the end of each cycle without rolling over.
For anyone who paid for the subscription and the API separately, part of the API bill has effectively moved inside the subscription fee. If you have wanted to try a small automation such as sorting work email or summarizing documents, pick one test before the credits expire. While credits remain, they are used before any credits you bought yourself.
These credits do not raise usage limits in ChatGPT or signed-in Codex, and they do not apply to Codex used through an API key. When they run out, API requests stop with an error, so automations tied to real work need a top-up method and an alert set up separately.
Pro 200 usage cut canceled, Codex limits doubled
In an October 29 notice, OpenAI said it is raising Work and Codex usage on Pro and Business Premium plans. Codex is OpenAI's coding assistant that writes and fixes code for you, and the notice treats ChatGPT's Work feature and Codex as one shared usage allowance. Pro 100 and Pro 500 now get twice their previous allowance, and the Pro 200 reduction announced by email on September 29 has been canceled.
Pro 200 subscribers who already had the higher allowance keep it, while those on the lower allowance get double their previous amount from October 29, bringing them to the same level. On October 26 this newsletter reported that the grace period for existing subscribers would end on the 29th with a smaller allowance; that plan was reversed on its final day.
People building or fixing internal tools with Codex can revisit the habit of splitting work to stay under the limit. Whether to move up a tier or stay put is best decided after looking at this week's actual usage share.
The allowance is only described in relative terms such as 'twice the previous amount', so how many more of your own tasks will fit has to be measured. How fast the limit drains also depends on task length and model settings.
Daishin Securities leak hits 38,490 amid finance hacks
On October 28 it emerged that data on 38,490 Daishin Securities customers had leaked. According to reports, a server linked to remote account opening was hacked from outside, exposing names, email addresses, home addresses, bank names, account numbers and mobile numbers (some account and phone numbers were encrypted). The company said it found no secondary damage and no access to its core trading systems.
Around the same time, Mirae Asset Global Investments confirmed abnormal access to some web systems, and Bithumb confirmed a leak of employee data through its internal mobile work system. Financial authorities said more than 4.4 million scanning attempts aimed at the financial sector had been detected since the 21st from a North Korean IP range (a block of internet addresses). Scanning is advance reconnaissance, knocking on doors one by one to find an open one; authorities said none of it has yet led to an actual breach and no direct link to the leaks above has been confirmed.
When names, contact details and banks leak together, convincing fake texts or calls about an 'account security check' become likely. If you are a customer of these firms, verify any individual notice through the official app or main phone number, and never log in through a link in a text message. Companies running servers connected to outside vendors should check the access logs at those connection points this week.
Investigations into how the leaks happened and how far they reach are ongoing, and who is behind the financial-sector hacks has not been established. Causes and any further damage should be judged from follow-up announcements.
Gemini 4 Carbon opens to security teams first
On October 28 Google announced Gemini 4 Carbon, describing it as its strongest model for cybersecurity and long-running tasks. Starting this week it is available only to a subset of cyber defenders, with other trusted partners added gradually through Fairwind, Google's early-access program. For developers, businesses and consumers Google said only 'as soon as possible', with no date.
The introductory price is $2 per million input tokens and $10 per million output tokens. A token is the unit AI uses to chop text into small pieces, roughly a cluster of a few characters. This matches Gemini 4 Argon's introductory price, and after the introductory period it doubles to $4 for input and $20 for output, at a time Google did not specify.
Because the model cannot be used yet, there is little reason to change work plans now. But if you are budgeting around the Gemini API, it is safer to check that the numbers still work at the doubled price, not just the introductory one.
The security scores Google highlighted were measured with safeguards turned off, so they are hard to read as the performance of the product ordinary users will get. Which apps and services it will reach has not been announced.
IeumOffice edits Hangul (HWP) files in their native format
IeumOffice is a desktop office program that opens and edits Hangul, Word, Excel, PowerPoint, PDF, Markdown, mind maps and more in a single window. It was built by a high-school Korean-language teacher on top of GenOffice and is open source (its source code, the blueprint, is public so anyone can read and modify it) under the Apache-2.0 license, with no usage fee. The current version is 0.12.8, and Windows 10/11, Mac and Linux builds are all in beta, the trial stage before a full release.
Hangul files are saved in their original format, .hwp as .hwp and .hwpx as .hwpx, and the first overwrite keeps the original as a .bak file. The developer says hundreds of real Hangul documents were checked by reopening them in Hancom Hangul 2024. Because AI tools often cannot handle Hangul documents directly, converting to another format and back, and breaking the layout along the way, has been common.
The AI panel is optional; linked to an existing Claude or ChatGPT subscription, it edits documents within that subscription's usage limit without a separate API key. It shows the AI's changes and lets you undo them, and the developer says it sends no usage statistics and opens and saves documents on your own PC. It suits professionals handling official letters, school notices and reports that must keep Hangul formatting.
Every platform is in beta, and the developer offers no warranty. Windows may show a warning because the program is new, and anything you ask the AI is sent to the connected AI company, so sensitive documents are better edited without the AI panel.
MacBook Pro with PROLED XDR: Korean pre-orders open today
Pre-orders in Korea for the MacBook Pro with PROLED XDR display, announced by Apple on October 28, opened on October 30, with deliveries starting November 6. It uses a tandem OLED screen that stacks two light-emitting OLED layers, and it is the first Mac to support touch on the screen. It comes with M5 Pro or M5 Max chips and up to 128GB of memory.
Korean starting prices are 4.79 million won for the 14-inch M5 Pro, 5.59 million won for the 16-inch M5 Pro, 7.59 million won for the 14-inch M5 Max and 8.09 million won for the 16-inch M5 Max. The 14-inch MacBook Pro with the M6 chip, announced a day earlier, has a Liquid Retina XDR screen, up to 32GB of memory and a US starting price of $1,999. If you want the new screen and touch, look at M5 Pro or Max; if price comes first, look at M6.
For anyone planning to run AI models on their own computer, unified memory (memory shared by the CPU and graphics chip) matters more than the screen. More memory lets you load larger models, so check how much memory your target model needs before pre-ordering.
Apple's press release did not give the time Korean pre-orders opened. How naturally touch works in existing Mac apps will only be clear after release.
Try this today
If you subscribe to ChatGPT Pro or Business, turn on the API credit benefit under Billing in settings and pick one automation to test within this billing cycle. Unused credits do not roll over.
If you use Codex, note today's usage share and compare next week how many more runs of the same task fit after the increase.
If you bank with Daishin Securities, Mirae Asset Global Investments or Bithumb, do not tap links in texts about security checks or refunds; verify through the official app or main phone number instead. Pass the same warning to family and staff.
If you edit Hangul documents often, install IeumOffice, open and save a copy of one file, then check in Hancom Hangul that the formatting is intact.