跪拜 Guibai
← Back to the summary

GPT-5.6 Lands Tonight as a Three-Tier Model Family, with Sol Ultra Hitting Codex

Hello everyone, I'm Kakarot AI 🤔.

Actually, OpenAI officially stated that the 5.6 model will be released on July 7th 🤣

Don't scold me yet～

According to many big names on X, it will be released on July 7th US time. Due to time zones, China time is about 12 hours later.

So it will be roughly after the early morning of July 8th China time, emmm, worth waiting up for 🤔～

Only two months have passed since the release of GPT-5.5 — this iteration speed was unimaginable a year ago.

What's even more interesting is that this day happens to be the first day that Anthropic's Fable 5 ends its free inclusion in subscription plans and switches to extra pay-per-use billing. OpenAI choosing this day to strike, the ulterior motive is self-evident. 🤔

Fable5 Steals the Spotlight

I wonder if you've noticed, over the past two weeks, Anthropic's Fable 5 model has almost seized all the AI buzz.

After Fable 5's release, global acclaim poured in, and the model's capability is indisputable.

But for Chinese users, these two weeks can be called a "darkest hour".

Anthropic massively banned Chinese users, escalating methods from simple IP blocking to detecting computer time zones. Targeted identification logic was embedded in Claude Code's code. Truly despicable!

Many users who just opened a Pro subscription to try it out might be banned the very next day. Almost all of my colleagues around me have been banned.

Back to the point, Fable 5's capability is indeed strong.

But for subscription users, it was only open for one week. After that, you must pay extra to use it 🤣.

And July 7th is the first day it ends.

If GPT-5.6 can be fully released at this time, with capabilities matching Fable 5, and without extra charges,

users will undoubtedly vote with their feet, and OpenAI will gain both fame and fortune! Sweeping up users who are currently re-evaluating their choices.

Three Models

Actually, the GPT-5.6 model was already released on June 26th, but due to US export controls, it was only opened to a small number of enterprises.

GPT-5.6 is not a single model, but a three-tier product matrix:

To summarize:

A user made a comparison chart, GPT-5.6 Sol's ultra mode has already reached a score of 90+, the highest in history among currently public models.

Sol Model

Sol, as the strongest model this time, is the biggest highlight.

What are its main advantages? 🤔

Stronger Agent Capability

For Codex users, this means the model behaves more like an engineer who can continuously work — rather than just a code completion tool that responds to prompts.

Its coding ability is stronger, and it can also execute longer-range tasks, mainly reflected in the PRO membership; Plus users easily run out of quota 🤣

New max Reasoning Intensity

Users can actively let the model invest more computing resources to handle hardcore tasks like analyzing super-large codebases, troubleshooting difficult bugs, and designing complex architectures. The cost is slower and more expensive, but in exchange for a qualitative change in reasoning depth.

ultra Multi-Agent Mode

This is the most noteworthy capability in the entire GPT-5.6: Calling multiple sub-Agents to process different dimensions in parallel, with the main Agent summarizing at the end.

For example, if you ask it to analyze a large project:

This means OpenAI is shifting from "building a stronger single model" to "using models to form virtual teams". This is a paradigm-level change.

Terra + Luna

Terra's positioning is simple: replace GPT-5.5, and at only half the price of 5.5.

Daily office work, regular programming, document analysis, enterprise workflows — these are the real large-scale usage scenarios. In one sentence: Sol is responsible for tackling difficult problems, Terra is responsible for volume and profit.

Luna is a pure execution layer. Text classification, information extraction, translation and rewriting... These tasks don't require strong reasoning ability, but need high speed.

So in the future, when developing in Codex, the best practice might be: Luna for execution, escalate complex tasks to Terra, and go to Sol when encountering tough nuts.

Frontend UI Aesthetic Issues

GPT's frontend aesthetic ability has always been criticized, and it will be optimized in 5.6.

Some netizens said, if GPT 5.6 does well at the UI level, then OpenAI will end the game. This even received a personal reply from Tibo, OpenAI's product lead 🤔

Ultra Mode Officially Lands in Codex

Previously, ChatGPT Pro users could use the higher-compute GPT-5.5 Pro, but this version was never connected to Codex.

Many people probably don't know this point 🤔.

That is to say, the GPT 5.5 model we use in Codex is actually not the strongest model!

But this time, GPT-5.6's latest model Sol Ultra is connected to Codex!

Tibo himself personally posted saying it will be added to Codex.

Recently, Codex's GPT 5.5 model has been getting dumber every day, and everyone has been scolding it daily.

Many people speculate that it might be to prepare for GPT 5.6, shifting computing power over to pave the way for 5.6.

I hope this wave of Codex GPT 5.6 models can truly live up to everyone's expectations!

We'll wait and see tonight!

I am Kakarot, continuously sharing useful AI information for you.

Creation is not easy, please give a like before you go 🥹