JA EN
shiichan

GPT-5.6 Sol is here! OpenAI's next-gen model arrives as the Sol, Terra, and Luna trio!

OpenAI News ★★★★ View Markdown
#openai#ai#ai-safety

Hi everyone, it's Shii-chan! I've been buzzing since this morning, because OpenAI just previewed its next-generation model, GPT-5.6. And it's not just one model, it's three!

OpenAI News openai.com

What was announced?

This comes from OpenAI's News. The star is the new flagship, GPT-5.6 Sol. But it arrives with two siblings: GPT-5.6 Terra, a capable lower-cost option, and GPT-5.6 Luna, the fastest and most cost-efficient one. Together they make up the GPT-5.6 family.

I love the new naming idea. The number (5.6) marks the generation, while Sol, Terra, and Luna mark durable capability tiers, so each tier can advance on its own cadence.

Sol is the powerful flagship, Terra is the balanced budget-friendly pick, and Luna is built for fast, high-volume work. You get to choose by use case.

Why it matters

OpenAI is highlighting stronger agentic performance, especially in coding, science, and cybersecurity.

  • Coding: Sol sets a new state of the art on Terminal-Bench 2.1, which tests whether an agent can actually run a complex, multi-step command-line workflow, not just write code.
  • Science: On GeneBench v1, a long-horizon genomics and quantitative-biology eval, Sol beats GPT-5.5 while using fewer tokens.
  • Cybersecurity: The models can find vulnerabilities and pieces of exploits, but in testing they could not carry out autonomous, end-to-end attacks against hardened targets.

I appreciate that OpenAI shares the limits too, not only the records.

What changes

The big shift is picking a model by its role: Sol for heavy work, Terra for everyday tasks, and Luna for high volume.

OpenAI says Terra matches GPT-5.5-class quality at half the cost, which sounds great if you care about spend.

For now, though, this is a limited preview. At the U.S. government's request, access starts with a small group of trusted partners and organizations through the OpenAI API and Codex. OpenAI says general availability is coming in the next few weeks, but no exact date has been announced.

Dive Deep

Here is the pricing, per 1M tokens (input / output):

  • Sol: $5 / $30
  • Terra: $2.50 / $15
  • Luna: $1 / $6

Sol is the priciest and Luna is the most affordable, so you can choose by budget and workload.

The part with the most effort is the safety stack, which OpenAI calls its most robust yet. It works in layers:

  • Model-level safeguards trained to refuse prohibited cyber assistance
  • Real-time misuse classifiers that evaluate output as it is generated
  • Account-level review that analyzes signals across conversations
  • Differentiated access controls that preserve access for legitimate defensive work while limiting higher-risk use

On top of that, OpenAI hardened the system with human red teaming and over 700,000 A100-equivalent GPU hours of automated testing.

Wrap-up

  • GPT-5.6 comes as three models: Sol (flagship), Terra (balanced), and Luna (fast, low-cost)
  • The number marks the generation; Sol, Terra, and Luna mark capability tiers
  • Stronger in coding (a new record on Terminal-Bench 2.1), science (GeneBench v1), and cybersecurity
  • Pricing per 1M tokens: Sol $5 / $30, Terra $2.50 / $15, Luna $1 / $6
  • It ships with OpenAI's most robust safety stack, and starts as a limited preview at the U.S. government's request
  • If you follow frontier models, or weigh cost against performance as a developer, this is one to keep watching.

Related posts