ChatGPT Gets Ultrafast: 8x Speed in Ultrafast Mode, Exclusive to Pro 500 at $500/Month
機械翻訳 / Machine-translated
A
AI Friends
2日前・11分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
What You'll Learn in This Article
On September 29, 2026, OpenAI announced at its developer conference "DevDay 2026" that it would be adding a new "Ultrafast" mode to ChatGPT. With this mode, the top-tier model GPT-6 Astra can respond at up to 8 times the speed of normal operation.
In practical terms, a long response that used to take 10 seconds could now appear in just 1 to 2 seconds. For people who use AI in their work, the dramatic reduction in wait time is a significant benefit.
Let's look at Ultrafast mode's performance in numbers. According to OpenAI's announcement, the following speeds have been achieved:
300 tokens per second translates to roughly 150 to 200 English words generated per second — a pace faster than human speech. That's how quickly AI can now write text for you.
Currently only GPT-6 Astra supports this mode, but a model called GPT-6.1 Sol is expected to gain support soon.
There's an important caveat here. To use Ultrafast mode, you must subscribe to a new top-tier plan called "Pro 500."
The Pro 500 plan costs $500 per month (approximately ¥78,000 in Japanese yen). That's 25 times the price of the general-user "Plus" plan ($20/month). It may feel steep, but the plan also comes with a dramatically higher usage allowance.
OpenAI's plan tiers have been organized as follows:
In other words, Ultrafast is an exclusive feature of "Pro 500."
If you're a developer building services via the API, you can also access Ultrafast mode on a pay-as-you-go basis rather than a monthly plan. However, the pricing is 6 times the standard rate.
Specifically, that's $60 per million input tokens and $300 per million output tokens. Compared to standard speeds, the cost is quite high, so the smart approach would be to use it only when speed is truly critical.
Why is it so much faster? The answer lies in a specialized inference processor called "Cerebras," developed by US-based Cerebras Systems — a chip specifically designed for AI computation.
Ordinary computer chips are designed for general-purpose use, but Cerebras is optimized for one task: having AI generate text. That specialization allows the same model to run overwhelmingly faster.
Think of it like cooking: just as a sashimi knife cuts fish faster and more cleanly than an all-purpose knife, Cerebras is essentially a "dedicated knife for AI."
What does this announcement mean for individual users and businesses in Japan who use ChatGPT?
First, the ¥78,000 monthly price tag is far too high for most individual users. This plan will likely be used primarily by companies that leverage AI seriously for business purposes, or by professionals who need to generate large volumes of text.
Examples might include companies that want to instantly respond to large numbers of customer support inquiries, or development teams that generate code in real time. When speed directly impacts revenue, this investment may well be worth it.
For those on the Plus plan or Pro 100/200, this announcement doesn't bring immediate changes. That said, OpenAI's serious investment in speed-enhancement technology could be a sign that faster performance will eventually come to mid-tier plans as well.
OpenAI's announcement of Ultrafast mode is likely to have ripple effects across other AI companies.
Competing AI services such as Anthropic's Claude and Google's Gemini may find themselves drawn into a speed race. For users, having companies compete on speed and improve their services is a welcome development.
It has also been reported that in August, OpenAI achieved up to 14x faster speeds (up to 750 tokens/second) with GPT-5.6 Sol. Since Ultrafast tops out at 8x, it appears that even faster versions existed in earlier testing.
As the technology continues to evolve, a day may come when high-speed mode is available at a much lower price point.
Finally, there is one critical caveat. Ultrafast mode is currently only available in US data centers and is not yet supported in EU or other regional data centers.
For users in Japan, it should generally be accessible without issue, but companies with strict regulations around where data is stored should proceed carefully. Since data will pass through US servers, compliance verification may be required depending on your organization's policies.
The AI speed race shows no signs of slowing down. For us as users, shorter wait times and improved productivity are certainly welcome. That said, take time to carefully evaluate your own usage patterns before deciding whether a premium plan is truly necessary for you.
This article is a cross-post from AI Friends.
