OpenAI is previewing GPT-5.6 as a three-model family — Sol, the new flagship; Terra, a cheaper everyday workhorse; and Luna, a fast budget tier. Sol sets a record on Terminal-Bench 2.1 and edges out Claude Mythos 5. Here's what shipped and how to get it.
OpenAI just split GPT-5.6 into three tiers
On June 27, 2026, OpenAI opened a limited preview of GPT-5.6 — not a single model, but a family of three. Sol is the new frontier flagship. Terra is a balanced model for efficient, everyday work. Luna is a fast, affordable option built for high-volume jobs. For now, all three are available only to a small group of trusted partners and organizations, through the OpenAI API and Codex.
Sol is the headline
OpenAI calls Sol its strongest model yet, with the biggest gains in agentic work across coding, biology, and cybersecurity. The standout number: on Terminal-Bench 2.1 — a benchmark that tests command-line workflows requiring planning, iteration, and tool coordination — Sol scores 91.9%, a new state of the art that edges out Anthropic's Claude Mythos 5. It also posts stronger results than GPT-5.5 on the GeneBench v1 biology benchmark, alongside general improvements in token efficiency.
Two new capabilities come with it: a "max reasoning effort" setting for the hardest problems, and an Ultra mode that spins up subagents for multi-agent workflows. On the plumbing side, developers get explicit cache breakpoints, a 30-minute minimum cache life, and a 90% discount on cached-input reads.
Terra and Luna: the same family, cheaper
The interesting move this round is at the bottom of the lineup. Terra matches GPT-5.5's performance at half the cost, and Luna pushes price down further for workloads where speed and volume matter more than frontier reasoning. The three-tier split lets teams pick a model per task instead of paying flagship rates for everything.
- Sol — $5 / million input tokens, $30 / million output (same price as GPT-5.5).
- Terra — $2.50 / million input, $15 / million output.
- Luna — $1 / million input, $6 / million output.
Safety came with the launch
OpenAI says GPT-5.6 ships with its most robust safety stack to date. All three models are classified as "High capability" for cybersecurity and biological/chemical risk under the company's Preparedness Framework. The team spent weeks finding weaknesses, pressure-testing the system, and hardening it against real-world attacks — red-teaming that OpenAI pegs at more than 700,000 A100-equivalent GPU hours — with strengthened protections for higher-risk activity, sensitive cyber requests, and repeated misuse.
How to get it
- Now: Sol, Terra, and Luna are in limited preview via the OpenAI API and Codex for select partners.
- Coming weeks: OpenAI plans broader general availability across ChatGPT, Codex, and the API.
- July: a Cerebras deployment is set to launch at up to 750 tokens per second.
The names are a little awkward, though
One small aside: the names land a bit awkwardly. Sol, Terra, and Luna are all heavily crypto-coded — Sol is Solana's ticker, and Terra and Luna belong to the ecosystem behind the May 2022 collapse that wiped out roughly $45 billion in a week. Names like these also get recycled constantly by memecoins and outright scam tokens, so seeing them badged on a frontier AI model is a little jarring. There's no blockchain connection here — but the association is hard to shake, and the crypto press ran "crypto frenzy" headlines within hours of the announcement.
The bigger picture
GPT-5.6 reads less like a single leap and more like a lineup decision. Sol takes the benchmark crown and the agentic-coding bragging rights, but Terra and Luna are the part most teams will feel: the same generation of capability spread across price points, so the cheap model is now genuinely good enough for a lot of real work. With general availability "in the coming weeks," the practical question isn't whether to use GPT-5.6 — it's which of the three you reach for.
Source: OpenAI — Previewing GPT-5.6 Sol
