Just now, GPT-6.1 Sol Ultrafast has arrived.

Today marks the fourth day of Tibo's "28 consecutive days of release or reset" streak, and the blockbuster news is here!

Yesterday, OpenAI decided to open GPT-6 to all users including free and paid users, and today it has launched a brand-new model the Ultrafast version of GPT-6.1 Sol! Combined with the updated model control system, it can realize real-time response to adjustments of user instructions.

Turbocharging

GPT-6.1 Sol itself was just released at DevDay on September 29, positioned as "capabilities close to Astra at one fifth of the price". The standard edition API is priced at $2 per million input tokens and $10 per million output tokens, which is already one of the most cost-effective cutting-edge models at present.

The Ultrafast mode directly boosts the speed by 8 times on this basis.

OpenAI describes it as "Near-Astra intelligence at up to 8x faster speeds than Sol Standard". In other words, its intelligence level is consistent with the standard version, still approaching the flagship Astra, but the speed is 8 times higher.

This is also OpenAI's second iteration of the "Ultrafast" concept. Earlier this year, they cooperated with Cerebras to launch the preview version of GPT-5.6 Sol Ultrafast, which claimed that the output speed was as high as 750 tokens/s, 14 times the standard speed. The 8x speed increase of GPT-6.1 Sol Ultrafast seems relatively conservative, but considering that the intelligence level of 6.1 Sol itself has far surpassed that of 5.6 Sol, achieving 8x acceleration on a higher capability baseline actually makes its value higher.

The significance of this speed for development scenarios is beyond doubt. Typical use cases given by OpenAI include online troubleshooting, real-time navigation applications for agents, and any real-time interaction scenarios where "every second is burning money".

In terms of API pricing, the Ultrafast mode is priced at $12 per million input tokens and $60 per million output tokens, which is exactly 6 times that of the standard version.

Dominik Kundel, Developer Relations Engineer at OpenAI, gave a more intuitive positioning: "Intelligence close to Astra level, 8 times the speed of Sol, and the cost is only 1.2 times that of Astra."

Compared with the standard API price of Astra ($10 per million inputs and $50 per million outputs), Ultrafast is indeed only slightly more expensive, but it brings a response speed far exceeding that of Astra. If the speed of Astra is not sufficient in your business scenario, Ultrafast provides an option to "pay a little more for a huge speed boost".

The capability parameters of Ultrafast are very attractive, but there is a key piece of information that is almost buried in the announcement.

On the Codex and ChatGPT Work sides, Ultrafast is only open to Pro 500 subscribers, eligible pay-as-you-go Enterprise users and Edu users.

Pro 500 is OpenAI's high-end subscription plan launched this year with a monthly fee of $500. This means that for ordinary ChatGPT Plus ($20/month) or even Pro ($200/month) users, if they want to use Ultrafast in ChatGPT or Codex, they have to pay extra, which has caused dissatisfaction among users.

Some developers said after the trial that Ultrafast is really surprisingly fast under the XHigh effort mode, but the token consumption speed is equally staggering. The 25x usage quota corresponding to the $500 monthly fee can "burn 1% in just a few minutes" under this mode. Some people also questioned the restriction logic itself: "If Ultrafast will consume my usage quota faster, why not let all paid users choose by themselves? I am willing to take the consequence of using up my quota faster."

Of course, the API side is not restricted by subscription levels. Developers can directly call the gpt-6.1-sol model and enable the Ultrafast mode, paying on a pay-as-you-go basis. This is not a high threshold for teams building agents and developing real-time applications.

The immediate value of GPT-6.1 Sol Ultrafast lies in that it provides a combination of "near-flagship intelligence + extreme speed" at the API level, without waiting for Astra's price reduction. For ordinary users, the experience improvement brought by instant response may be more direct. After all, no one likes to watch the model run all the way in the wrong direction.

What will Day 5 bring? See you tomorrow.

This article is from the WeChat official account "Machine Heart", author: Machine Heart, editor: Leng Mao, published with authorization from 36Kr.