GPT-5.6 Sol first batch of internal test results are here! The cost for the same task is only half of Fable 5.

GPT-5.6 Sol preview version has been out for almost half a month, and the first batch of user beta test reports are finally fresh off the press!

NVIDIA's Chief Engineer puts it in the straightest, most no-nonsense terms:

Sol is a beast! In 30 hours, it outperformed the CUDA acceleration effect that Opus achieved in 64 hours.

After optimization in subsequent versions, it might completely crush Opus……

Netizens haven't been idle either, directly jumping into action with all sorts of comparisons (and digs) popping up:

Take building a spaceship for example—look at the interior corridor on GPT-5.6's side: the color matching and lighting choices exude a strong tech vibe, with clear layers of light and shadow. In contrast, on the 5.5 side, the hues are warmer and grayer, the overall image is much flatter, and it looks more like a company locker room.

Not to mention the space scenes outside the ship—GPT-5.5 is practically a mosaic party. (doge)

This round, GPT-5.6 wins decisively!

So why has GPT-5.6 become a hot topic again recently? First, the old rival Fable 5 has made a comeback; second, the model has fi!nal!ly! moved!

After waiting for so many days, news has finally spread that the model will officially launch within the next couple of days.

It won't just be available to a handful of "partners" this time—everyone gets it!

(Those who waited win~)

Altman also personally stepped in to flex, bringing the hype to a climax:

The moment my baby said his first two-word phrase was as mind-blowing as when GPT-5.6 discovered a brand new mathematical theory.

No lies! Both came from the same family (doge)

First Batch of Beta Test Results Released

Scrolling through beta testers' posts, a few common points stand out.

First, as the aforementioned NVIDIA engineer mentioned: cleaner code, overall fewer lines of code.

For the same requirement, other models might write a whole bunch, but Sol is noticeably restrained—if it can solve it in three lines, it won't write five. The code lines are only 1/5 of Opus. Especially with C++, the writing style is very close to real human coding, with fewer comments.

For projects that require long-term maintenance, Sol has the advantage.

Of course, coding isn't entirely without flaws—for example, iteration and advancement are slower, and failure rates are higher, because Sol always tends to attempt more difficult tasks.

Compared to Opus, its trial-and-error exploration approach is also more limited—once it locks onto a direction, it stubbornly sticks to it until the end.

In simple terms, the model has abandoned a large amount of mindless trial and error in favor of long-term deep cultivation. It doesn't chase superficial output quality but focuses on underlying performance optimization.

This is very much in line with OpenAI's positioning for Sol—

Designed for long-chain tasks such as high-difficulty reasoning and complex code, especially suitable for complex workflows that require planning, iteration, tool invocation, and step coordination.

Throwing the same prompt to Sol and GPT-5.5 Pro gives a very intuitive comparison:

Whether it's interactive SVG, 3D models, or game generation, Sol outperforms in instruction following and spatial reasoning, with better consistency.

In the area of front-end design, Sol has also delivered a cleaner and neater result.

Compared to GPT-5.5, Sol's page layout, hierarchy, and white space are well handled, and the visuals are more polished. Visually, Sol easily passes with flying colors.

Successfully Targeting Fable 5?

But the question netizens care most about—how does it compare to Fable 5?

The result may still be slightly inferior.

In some benchmark tests, Sol scores on par with or even surpasses Fable 5, but there is still a gap with Fable 5 in overall model experience and code quality.

For example, user Gipp did a 3D FPS game comparison.

While GPT-5.6 is still struggling to adapt to the game world, lighting, and gameplay, Fable 5 can already turn a single prompt into a fully playable finished game.

However, Fable 5's cost is much higher: $10 per million input tokens and $50 per million output tokens. Sol's input cost is only $5 per million tokens, with an output cost of $30.

Sol is almost half the price!

So with similar model capabilities, it's hard to say whether Sol or Fable 5 will come out on top.

More critically, safety restrictions. Ever since Fable 5 was sanctioned, after its return, netizens clearly feel Fable 5 isn't doing any work.

Normal code writing or debugging—one slip and the system flags it as high risk, and the task gets downgraded directly to Opus 4.8 for processing. Even questions like "how many 'r's are in 'raspberry'" are intercepted.

Netizens have given Fable 5 a thumbs down…

On the GPT-5.6 side, the situation is slightly better. The company has explicitly stated that stronger safety systems have been added, and different protection strategies will be configured based on model capabilities, with restrictions likely fewer than those on Fable 5.

In any case, we're eagerly waiting for GPT-5.6's full rollout. We'll see the real results when we test it ourselves then~

Source: QuantumBit

Risk Warning and Disclaimer

        Markets carry risks; invest with caution. This article does not constitute personal investment advice and does not consider the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article suit their particular circumstances. Any investment decisions based on this article are at your own risk.