OpenAI woke up on Thursday and chose absolute chaos, (the good kind, the "let's ship a whole family of models" kind).
The tech giant just dropped an entire family of models called GPT-5.6, and the branding is giving pure, unadulterated cosmic energy. We’re getting three distinct models: Sol (the strong, do-everything one), Terra (the reliable middle child), and Luna (the budget-friendly baby of the group).
Think of it like your morning caffeine routine. Sol is your fancy, quadruple-shot oat milk latte; Terra is the reliable, everyday cappuccino; and Luna is the cheap drip coffee that still gets the job done when you are desperate.
So how do they stack up?
OpenAI is positioning this launch to completely steal Anthropic's thunder. They went straight to the Artificial Analysis Coding Agent Index to prove that their shiny new flagship model, Sol, is outrunning the competition.
OpenAI is calling Sol its absolute best coding model to date. They even dragged Anthropic’s highly hyped Fable 5 model into the mud for a direct comparison.
According to the benchmark data, OpenAI's Sol achieves the following:
Higher Scores: It hits a state-of-the-art coding score of 80, sitting exactly 2.8 points above Fable 5.
Token Efficiency: It accomplishes this while chewing through less than half the output tokens. In fact, CEO Sam Altman told CNBC that Sol is a whopping 54% more token-efficient for coding tasks than previous models.
Insane Speed: It processes the tasks in less than half the physical time.
Budget Friendly: It costs roughly one-third less than Anthropic's alternative. Talk about a public callout.
But that’s not all:
The entire GPT-5.6 family is built to flex across enterprise workflows, heavy programming loops, and deep scientific research. OpenAI is heavily bragging about its defensive capabilities, calling this their strongest cybersecurity package yet.
The software is explicitly tuned for hardcore defensive tasks like threat modeling, scanning and patching vulnerable code, and blue teaming, which is essentially white-hat hacking your own infrastructure before the bad guys get a chance to do it.
Alongside the models, OpenAI unveiled a workspace assistant called ChatGPT Work. Built specifically for corporate teams, this desktop, web, and mobile application is designed to help professionals generate documents, financial spreadsheets, and interactive presentations instantly. Check out the official OpenAI ChatGPT Work Product Console for integration steps.
For The Pricing and Tiers Breakdown:
The update is officially live and rolling out across ChatGPT, Codex, and the OpenAI API infrastructure.
Here’s exactly how the pricing shakes out per million tokens:
Sol (The Flagship): $5.00 input / $30.00 output. Built to crush Anthropic's high-tier models.
Terra (The Middle Child): $2.50 input / $15.00 output. Optimized to run just slightly above Fable 5 capabilities.
Luna (The Budget Baby): $1.00 input / $6.00 output. OpenAI claims this lightweight model still comfortably outperforms Anthropic's Opus 4.8.
Now for the real behind-the-scenes tea: This rollout almost did not happen. Washington insiders previously revealed that the Trump administration actively attempted to stall or restrict the deployment of GPT-5.6.
Government officials raised major red flags regarding its advanced cyber warfare capabilities and potential for widespread misuse. While the models are officially live today, it remains unconfirmed whether OpenAI completely cleared those regulatory hurdles or if they’re currently playing a dangerous game of chicken with federal watchdogs.
As for those misuse concerns? Only time (and real-world use) will tell if the new models actually warrant the worry.
To read the full technical and political breakdown of the launch, check out the original report on OpenAI blog post, and TechCrunch.
