OpenAI Unveils GPT-5.6, a Three-Tier Model Family Balancing Frontier Intelligence and Efficiency
OpenAI has announced GPT-5.6, its next-generation model family. It ships in three tiers — the flagship “GPT-5.6 Sol,” the balanced everyday model “GPT-5.6 Terra,” and the fast, low-cost “GPT-5.6 Luna” — with efficiency gains per token that push frontier-level intelligence and cost efficiency forward at the same time. The family is designed with agentic workloads like Codex and ChatGPT Work in mind.
Details
- Three-tier lineup: Sol offers maximum reasoning capability, Terra is a mid-tier model for everyday use, and Luna is the fastest and most affordable of the three
- Benchmarks: Sol beats Claude Fable 5 on the Artificial Analysis Coding Agent Index. Terra matches the intelligence-benchmark scores of the previous model, GPT-5.5, at a lower price
- Pricing: Sol’s cost for coding tasks is less than half that of Claude Fable 5. Terra delivers GPT-5.5-level intelligence at 50% of GPT-5.5’s price. Luna costs 80% less than Sol
- Efficiency gains: Improvements to speculative decoding increased token-generation efficiency by more than 15%, while kernel optimizations cut end-to-end serving costs by 20%. The models themselves were trained to “get more work done per token”
- Self-optimization: Using kernel languages like Triton and Gluon, the model can autonomously optimize production kernels. Correctness is validated with a proprietary tool called FpSan (Floating-Point Sanitizer)
- Scale: ChatGPT serves 1 billion active users globally and more than 2 million businesses; GPT-5.6 was designed with a four-year deployment horizon in mind
- Intended use: Built around agentic products like Codex and ChatGPT Work, leveraging speculative decoding, KV cache optimization, and prompt caching
How to try it
- Select a model in the “GPT-5.6” family (Sol, Terra, or Luna) from the model picker in ChatGPT, ChatGPT Work, Codex, and other supported products
- Choose Sol for maximum performance on coding-agent workloads, Terra for everyday use with balanced cost, or Luna for lightweight, low-cost tasks
- Detailed API rollout timing and model names are expected to be announced progressively