GPT-5.6 Is Here: Three Models, New Pricing, and a Political Twist
OpenAI launched GPT-5.6 with three tiers: Sol, Terra, and Luna. Here's what the pricing, capabilities, and Trump administration delay mean for your work.
LUMIEN4 min read
OpenAI launched GPT-5.6 on Friday as a limited preview, less than a day after reports emerged that the Trump administration had asked the company to hold off on the release. The new suite includes three models: Sol, the flagship; Terra, aimed at high-volume work; and Luna, a fast and lower-cost option for everyday use. OpenAI says the models are particularly strong at coding, cybersecurity, biology, and staying on track through long agentic tasks.
What happened
On Friday, OpenAI released GPT-5.6 as a limited preview. The timing raised immediate eyebrows: news had broken less than 24 hours earlier that the Trump administration had asked OpenAI to delay the release. OpenAI went ahead anyway.
The suite ships as three distinct models, each aimed at a different use case:
- Sol: The flagship model, built for the most demanding tasks.
- Terra: A mid-tier option described by OpenAI as suited for “high-volume work.”
- Luna: The lightweight tier, positioned as fast and affordable for everyday use.
According to OpenAI, the suite performs especially well at coding, cybersecurity, and biology. The company also claims it handles long-horizon agentic tasks better than previous models, meaning it stays focused and accurate across multi-step workflows without losing the plot halfway through.
How the pricing stacks up
Pricing for GPT-5.6 Sol comes in at $5 per million input tokens and $30 per million output tokens. Compare that to Anthropic’s Claude Fable 5, which is priced at $10 input and $5 output per million tokens.
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| GPT-5.6 Sol | $5 | $30 |
| Anthropic Claude Fable 5 | $10 | $5 |
The difference in structure matters depending on your workload. Sol costs less to feed data into, but its output cost is six times higher than Claude Fable 5. If you run tasks that generate long outputs (code, reports, summaries), that $30 output rate adds up quickly. Claude Fable 5 flips the cost profile: you pay more to send information in, but get responses back cheaply. Applications that process large inputs and return short answers will likely do better on Claude’s pricing.
Why it matters
The political angle is notable. If the Trump administration’s reported request was real, OpenAI’s decision to release within 24 hours signals that the company is not willing to let government timing preferences dictate its product roadmap, at least not for long.
On the product side, the three-tier structure is becoming the industry standard. OpenAI, Anthropic, and Google all now ship model families rather than single flagship releases. That gives developers more flexibility but also more decisions to make: which tier fits which task, and how do the token costs interact with your actual usage patterns?
The claimed strengths in coding, cybersecurity, and biology suggest OpenAI is targeting professional and enterprise buyers directly, not just general consumer use. Long-horizon agentic performance is also a real battleground right now. Agents that wander off course mid-task are one of the biggest friction points for teams trying to automate multi-step workflows.
Our take
The pricing table above is the most useful thing to look at before you do anything else. Sol’s output cost being six times higher than Claude Fable 5’s is not a minor footnote. If your application writes a lot of tokens out (think code generation, document drafting, or detailed analysis), that gap will show up in your bill fast.
The agentic task claims are worth testing but not worth trusting at face value yet. Every major lab says its latest model is better at staying on track through complex workflows. The proof is in running your actual tasks, not reading the announcement.
The three-model structure (Sol, Terra, Luna) is sensible, but it also means you need a clear routing strategy. Defaulting to Sol for everything because it is the flagship is an easy way to overspend. Terra and Luna exist for a reason, and figuring out where they fit in your stack should be an early priority once you get access.
The Trump administration subplot is interesting but practically irrelevant for most operators. What matters is whether the model performs, at what cost, and whether the limited preview becomes general availability soon.
What to do about it
Request access to the limited preview if you are actively building on the OpenAI API. Run your two or three most token-heavy workflows through Sol and compare the output quality and cost against your current model. Pay specific attention to tasks where the output is long. That $30 per million output rate is where the real cost difference lives, and a quick test will tell you whether Sol’s quality justifies it for your use case or whether a cheaper tier (or a competitor) is the smarter call.
