OpenAI unveils GPT-5.6 amid US AI regulatory drama

OpenAI dropped a limited preview of its next‑generation GPT‑5.6 model suite on Friday, a move that comes just hours after reports that the Trump administration had asked the company to stagger its rollout. According to The Verge, the announcement includes three distinct variants: Sol as the flagship, Terra aimed at high‑volume workloads, and Luna positioned as a fast, affordable everyday option.

The three models are marketed as experts in coding, cybersecurity and biology, and the company claims they can maintain focus on long‑horizon agentic tasks. That focus on specialized competence hints at a broader shift toward domain‑specific AI, something many enterprises have been asking for as general‑purpose models hit diminishing returns.

Pricing is where the headlines get louder. OpenAI says Sol will cost $5 for input tokens and $30 for output per million tokens, a figure that is roughly half the cost of Anthropic’s Claude Fable 5, which the report lists at $10 for input and $5 for output. If those numbers hold up, OpenAI could be setting a new price benchmark for high‑performing models.

Regulatory pressure appears to be reshaping OpenAI’s release strategy. By offering a limited preview rather than a full launch, the firm can gather real‑world feedback while staying within the bounds of the administration’s request. This approach mirrors how other tech giants are testing features in sandbox environments to avoid compliance pitfalls.

From a developer’s perspective, the tiered lineup gives more flexibility. Teams that need raw power can dip into Sol for complex bio‑informatics queries, while cost‑sensitive operations might stick with Luna for routine customer‑support automation.

What this means for you: if you’re already using GPT‑4 for code generation, you can experiment with the new Sol model to see if the higher output cost translates into faster, more accurate results. Try this prompt with an AI assistant: "Write a Python script that scans a directory for vulnerable dependencies, flags them, and suggests secure alternatives, using the GPT‑5.6 Sol model for detailed analysis." This workflow can help you benchmark performance and cost before committing to a full migration.