OpenAI launches GPT‑5.6 Sol, Terra and Luna for reasoning and coding

OpenAI today unveiled three GPT‑5.6 models—Sol, Terra and Luna—each built for a different slice of the professional market. Sol is positioned as the flagship for complex, high‑stakes tasks, while Terra promises a middle ground of intelligence and affordability, and Luna targets high‑volume, cost‑sensitive workloads. The release marks OpenAI’s first multi‑model rollout since the controversial pause on GPT‑6 Astra, a move that underscores the company’s tightening safety controls.

Sol targets complex professional work with a 1.05M token context window and a knowledge cutoff of February 16, 2026. It costs $4 per input token and $20 per output token, reflecting its premium positioning. According to developers.openai.com, Sol supports functions, web search, file search and computer use, delivering reasoning scores that sit above the previous GPT‑5 family. Early adopters report faster computer‑use tasks and more reliable code generation for large‑scale projects.

Terra is billed as the balanced option, offering a 1.05M context window at a $2 input price and $12 output price. Its reasoning profile is lower than Sol’s but still exceeds the baseline, making it suitable for teams that need stronger AI capabilities without the $4‑per‑token sticker shock. The model’s knowledge cutoff matches Sol’s, and it inherits the same tool suite, which includes web search and file search. For organizations looking to upgrade from GPT‑4 without a massive budget, Terra provides a clear upgrade path.

Luna is the most economical of the trio, charging just $0.20 per input token and $1.20 per output token. It retains the 1.05M context window and February 16 2026 knowledge cutoff but scales back reasoning to the lowest tier among the GPT‑5.6 family. According to cryptobriefing.com, Luna is optimized for high‑volume, cost‑sensitive workloads such as bulk content generation or routine code refactoring. Its tool support mirrors the other models, but performance benchmarks show it trails Sol and Terra on complex reasoning tasks.

All three models share a common architecture: they accept text and image inputs, produce text outputs, and support multilingual prompts. They are accessible via the Responses API and OpenAI’s client SDKs, giving developers a unified interface for switching between tiers. The pricing structure lets teams experiment with Sol’s advanced reasoning, graduate to Terra for daily operations, and finally offload routine tasks to Luna to keep costs low.

The market reaction has been mixed. While some engineering managers praise the granularity of the new pricing tiers, others point to the broader industry context. cnet.com notes that the launch comes as OpenAI paused reinforcement‑learning work on GPT‑6 Astra after it triggered the company’s “critical” cybersecurity threshold. The pause, announced by Sam Altman on August 18, 2026, was intended to create isolated testing environments for the model, which scored 100 % on ExploitBench and 98 % on FrontierMath Tier 4. The decision highlights OpenAI’s shift toward more cautious deployment as it rolls out the GPT‑5.6 family.

For developers, the new lineup signals a maturing of OpenAI’s product strategy. Instead of a single monolithic model, the company now offers tiered options that align with the principles outlined in “artificial intelligence basics” and “artificial intelligence: a modern approach”—choosing the right tool for the job reduces waste and improves safety. The availability of a low‑cost model like Luna also opens doors for smaller firms and research labs that previously could not justify premium pricing. Moreover, the emphasis on reasoning and coding reflects the growing demand for AI that can handle technical documentation, a gap highlighted by recent open‑source diagram engines such as usatoday.com’s Schematex, which struggles with spatial reasoning that GPT‑5.6 models aim to address.

Will enterprises adopt a multi‑model strategy, or will the complexity of managing three distinct APIs outweigh the cost benefits? The answer will shape how artificial intelligence evolves in professional settings over the next year.

FAQ
What are the key differences between GPT‑5.6 Sol, Terra and Luna?
How does the pricing of the new models compare to previous GPT generations?
Why did OpenAI pause work on GPT‑6 Astra?
What tools and capabilities do the GPT‑5.6 models support?