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 generation's capabilities at GPT-5.5's price point.
That's a key point.
Because Claude's top-tier Fable 5 / Mythos 5 official pricing is:
Input: $10
Output: $50
GPT-5.6 Sol is effectively 50% cheaper on input and 40% cheaper on output.
But if it were just cheaper, that wouldn't be ruthless enough.
What's even more ruthless is that OpenAI mentioned Sol can compete with Mythos Preview on ExploitBench, while using only about 1/3 the output tokens.
What does this mean?
It's not simply "cheaper per token."
It means for the same complex tasks, it likely has less fluff, shorter paths, and a lower total bill.
That's the real blow aimed at Claude.
Claude has always been strong at long tasks, coding, and agent-like behavior, especially with Claude Code's solid experience.
But GPT-5.6 Sol isn't striking a single parameter; it's a combination punch:
Price doesn't increase
Speed is faster
Output is more efficient
Coding continues to top the charts
And Cerebras support is on the way, with up to 750 tokens/s
A speed of 750 tokens/s, if truly applied in Codex/Agent scenarios, would be an incredibly dramatic experience.
In the past, we judged a model's strength mainly by whether it answered correctly.
But in real coding scenarios now, strength also depends on:
Can it take fewer detours?
Can it use fewer words?
Can it run fast?
Can it keep costs under control?
Can it handle long tasks without breaking down?
From this perspective, GPT-5.6 Sol isn't really pitted against GPT-5.5.
Its true target is Claude's highest-tier model.
My assessment:
GPT-5.5 already elevated Codex's coding experience to the top tier.
If GPT-5.6 Sol can simultaneously deliver on price, speed, token efficiency, and coding capability, then Claude Code's biggest moat today will be significantly weakened.
In particular, the Terra tier, I think, will be terrifying.
It's not the flagship, but it's officially said to be close to GPT-5.5, yet at half the price.
For a large number of developers, that might be more devastating than Sol.
Sol's job is to prove that OpenAI can still compete at the top capability level.
Terra's job is to commoditize GPT-5.5-level capability.
Luna's job is high-speed, high-volume.
With this lineup, Claude will be under tremendous pressure.
So right now, what I'm most looking forward to isn't how GPT-5.6 performs in ChatGPT.
I only care about one thing:
Once Codex integrates GPT-5.6 Sol, in large real-world projects, can it actually take down Claude Code's advantage?