OpenAI just dropped its newest family of models. On Thursday the company unveiled GPT-5.6, a set of three heavyweight variants aimed squarely at enterprise work, coding, and cybersecurity, according to TechCrunch AI. This is a big one, and the marketing makes no secret of the target: Anthropic.
Here’s what shipped and why it matters.
Three models, three price points
GPT-5.6 comes in three flavors, each tuned for a different job and budget:
- Sol is the workhorse. OpenAI calls it its best coding model yet.
- Terra sits in the middle as a balanced option.
- Luna is the budget pick for lighter tasks.
Pricing per million tokens breaks down like this:
- Sol: $5 input / $30 output
- Terra: $2.50 input / $15 output
- Luna: $1 input / $6 output
CEO Sam Altman is leaning hard on efficiency. He recently told CNBC that Sol is 54% more token efficient on coding tasks than previous versions. That number matters more than it looks. Token efficiency is the difference between a model that’s impressive in a demo and one a company can actually afford to run at scale.
The cybersecurity angle
OpenAI calls 5.6 its “strongest cybersecurity model yet, achieving frontier performance with significantly fewer tokens.” That claim carries weight, because TechCrunch AI reports the Trump administration previously tried to restrict the model’s rollout over fears it could be misused.
The company is framing 5.6 as a defensive tool. It supports:
- Threat modeling
- Code review and patching
- Blue teaming, which means simulating an attack on your own systems to find the holes before real hackers do
What stands out here is the tension. A model strong enough to raise regulatory eyebrows is also being pitched as the thing that keeps your systems safe. Both can be true, and that’s exactly why this space is getting harder to govern.
ChatGPT Work joins the lineup
Alongside the models, OpenAI released ChatGPT Work, a workplace companion for enterprise teams. It runs on desktop, web, and mobile, and handles the daily clerical grind: drafting documents, building spreadsheets, and putting together presentations. It’s a direct play for the office worker’s day, and a signal that OpenAI wants to live inside the workflow, not just answer questions on the side.
The real fight is with Anthropic
GPT-5.6 follows similar releases this week from SpaceXAI and Meta. But TechCrunch AI notes the whole rollout looks built to counter Anthropic, which has quietly become the likable underdog of the AI race by focusing on enterprise customers.
OpenAI isn’t being subtle. Citing the Artificial Analysis Coding Agent Index, a respected benchmark, it claims the new family beats Anthropic’s models across the board. The specific shot at Anthropic’s much-hyped Fable:
- Sol sets a new state of the art at 80, 2.8 points above Fable 5, while using less than half the output tokens, taking less than half the time, and costing about one-third less.
- Terra performs just above Fable 5.
- Luna outperforms Opus 4.8.
Benchmarks are marketing until independent testers confirm them, so treat those figures as OpenAI’s own scorecard for now. Still, the message is clear. OpenAI wants to win on speed and cost, not just raw capability.
What it means for you
If you build with these models, the practical takeaways are simple:
- Cost per task is the new battleground. Fewer tokens for the same output means lower bills, which is what actually moves enterprise buyers.
- You’ve got real tiering now. Route cheap work to Luna, heavy coding to Sol, and stop overpaying for tasks that don’t need the top model.
- Verify the benchmarks yourself. Run your own evals before you migrate anything important. Vendor scorecards rarely survive contact with your real workload.
GPT-5.6 is live now across ChatGPT, Codex, and the OpenAI API. The next move belongs to Anthropic, and given how this week has gone, the response probably won’t take long. You can find the full details at the original source.