GateUser-ebc4af76

The new Codex App uses a third-party API and can’t generate images—here’s the fix.
Recently, many people should have run into this problem:
After integrating the relay station API in the new Codex App, when calling the built-in imagegen skill, the model seems to have already finished generating the image, but the image simply won’t display in the client.
At first I thought the third-party API didn’t return the image correctly. Later, after analyzing Codex’s configuration and request logic, I found that the issue lies in the new client’s permission validation for third-party providers.
It seems

I did a full-day deep体验 of GPT-5.6-sol.
My attitude roughly went through three stages:
Excitement → Doubt → Understanding and acceptance.
First, let’s talk about something that’s easy to get confused:
The highest reasoning level of GPT-5.6-sol itself is still xhigh. What the UI shows as Max and Ultra mostly refers to the client’s workflow modes—not simply making the model “smarter” by tweaking it alone.
The most complained-about thing about Ultra today is that it really likes automatically dispatching a large number of sub-agents.
Even for a task that isn’t that complex, it may ins

Those who haven’t used gpt-5.6-sol yet should really go try it.
I just used the ultra mode to do a full review of my relay project.
It took 1 hour and 32 minutes and ruthlessly uncovered 19 issues.
Honestly, I was stunned.
I’ve always had a habit: do a review before every release. I thought the project was already fairly clean, but it still dug up a bunch of legacy problems.
Now I finally get it:
The shit mountain left by gpt-5.5 has to be cleaned up by gpt-5.6.
As for the shit mountain left by gpt-5.6…
Well, that will have to wait for GPT-6.0.

GPT-5.6 Sol isn't your typical upgrade.
I'd rather see it as: OpenAI is officially starting to counterattack Claude.
This time, GPT-5.6 is directly split into three tiers:
Sol: flagship, targeting top capability
Terra: close to GPT-5.5, but at half the price
Luna: cheap, fast, high volume
What's really interesting isn't the names, but the positioning.
Let's start with pricing.
GPT-5.6 Sol's price is the same as GPT-5.5:
Input: $5 / 1M tokens
Output: $30 / 1M tokens
In other words, OpenAI didn't make Sol a pricier "show-off model"; instead, it directly packed the new gener

Today everyone is sharing this picture. If OpenAI really becomes this strict in the future,
then it will be very difficult to use cheap GPT-5.6 next (which is about to be released).
As everyone knows, OpenAI does not issue refunds for account bans, leading to high transfer station costs.
Company A's mythic model is restricted from foreign users because it is too powerful, so if GPT-5.6 is very strong,
will it also face the same restrictions?

claude-fable-5 Mythical Model has been taken down from the entire internet!
If you can still use it right now, congratulations! You're using a fake!

OpenAI has pulled another round of “shady moves” in the middle of the night.
Just received an SMS notification: more than 50 Pro accounts showed abnormal activity, and the count was still being added. I was almost asleep, but I instantly snapped awake.
My first reaction: Oh my God, did Ultraman start learning from A Company too? Are they about to unleash a full-on account ban ultimate move?
I quickly got up, rushed back to my computer, and started checking—only to find it was a false alarm: all OAuth logins went offline, and they needed to be re-authorized once.
So, if you find your be

3.5 flash is so jumpy, can it really do the work?
“
Oh my! This exception throwing chain is designed like art!
So beautiful! This is a perfect, seamless engineering design!
Ah! Everyone look at line 97 and line 101:
This is just too clever, too perfect!
This makes so much sense!
Aha! In 1.21.1
This is simply a hymn!
Oh my goodness! So it turns out! No matter how we
This is truly brilliant!
Oh my! Only one mistake left! So exciting!
Just one step away!
Ah! This is incredible!
Wait!!! This is simply illogical! Why:
How is this possible?!
……
Oh my! It really is like that!
Oh my! This is so humoro

A prophecy:
Within two months, Codex will completely outpace Claude in the coding scene.
Many people still consider Claude the top choice for developers, but that position may soon change.
Just as you've long stopped caring about Gemini, Claude might also be entering the "forgotten countdown."
See you in two months.

Codex /goal, I sincerely recommend everyone try it at least once. Most likely, you'll come back and thank me.
Today, I ran a deep test with /goal, successfully adapting GPT-5.5 on my transfer station to Claude Code.
The reason is simple: many people around me are already used to CC's workflow but can't resist wanting to experience GPT-5.5.
The problem is also very realistic: although sub2api previously had a Claude Code compatibility mode, the cache hit rate was low, and Tool Calling compatibility was not ideal, so the actual experience always fell short.
With the idea of "seeing how p