GPT-5.6 Sol vs Terra vs Luna: Which Tier Is Right for You?
OpenAI released a limited preview of the GPT-5.6 model family on June 26, 2026, introducing a new naming system alongside three distinct capability tiers: Sol, Terra, and Luna. The models are not yet publicly available; access is restricted to roughly 20 vetted organizations via API and Codex. The pricing and positioning are already clear enough to map against real use cases.
This post covers what each tier does, who it's built for, what it costs, and what we still don't know.
The New Naming Convention
GPT-5.6 marks a shift in how OpenAI labels its models. The number (5.6) identifies the generation. The name (Sol, Terra, or Luna) identifies a durable capability tier that can be updated on its own schedule, independent of the generation number.
In practice, that means future improvements to, say, Terra don't require a whole new generation number. Each tier can advance without disrupting the others. OpenAI is signaling that this naming system is meant to stick around.
The Three Tiers
GPT-5.6 Sol: The Frontier Tier
Sol is OpenAI's most capable model in this family. It's designed for work where correctness and depth matter more than speed or cost: multi-step coding projects, extended research tasks, vulnerability analysis, and agent-driven workflows that need to sustain context across many steps.
The standout benchmark OpenAI cited is Terminal-Bench 2.1, where Sol hit 91.91% in ultra mode. OpenAI also claims Sol outperforms GPT-5.5 on long-horizon genomics analysis. Whether those numbers hold up across a wider evaluation suite will become clearer at general availability, when OpenAI has promised a fuller benchmark disclosure.
Sol introduces two new reasoning modes:
- max effort: extends depth for a single problem-solving chain, prioritizing correctness over latency
- ultra mode: deploys parallel subagents to work on different parts of a complex task simultaneously, trading speed for accuracy
Pricing for Sol is $5 per million input tokens / $30 per million output tokens, the same as GPT-5.5. Sol also carries the most layered safety architecture of the three: model-level refusals, real-time output classifiers, account-level review, and over 700,000 A100-equivalent GPU hours of automated red-teaming. OpenAI notes Sol does not cross the "Cyber Critical" threshold but may produce false positives on legitimate security research during the preview window.
Who Sol is for: Teams running agentic pipelines, security researchers, anyone doing long-horizon scientific or coding work where a wrong answer is more expensive than a slow one.
GPT-5.6 Terra: The Production Tier
Terra is positioned as the balanced option: performance comparable to GPT-5.5 at roughly half the cost. OpenAI describes it as the primary migration target for existing GPT-5.5 API users.
Pricing: $2.50 per million input tokens / $15 per million output tokens, approximately 50% cheaper than Sol and GPT-5.5.
Terra supports the same prompt caching improvements as the rest of the family: a 90% discount on cache reads, cache writes billed at 1.25x the uncached input rate, and guaranteed 30-minute minimum cache retention with explicit breakpoints. For high-volume production use, those caching mechanics matter as much as the base rate.
Who Terra is for: Product teams running production workloads - customer-facing AI features, document processing, content pipelines - where you need GPT-5.5-level quality but the cost of running Sol at scale doesn't pencil out.
GPT-5.6 Luna: The High-Volume Tier
Luna is built for tasks where you're processing a lot and speed matters: classification, extraction, routing, first-pass drafting, summarization at scale. It's the cheapest of the three and the fastest.
Pricing: $1 per million input tokens / $6 per million output tokens.
OpenAI notes that Luna's performance on security-related tasks improves with higher reasoning budgets, meaning you can push Luna harder on structured tasks if the latency trade-off is acceptable.
Who Luna is for: Developers building high-throughput pipelines - data enrichment, content triage, bulk classification - where you need throughput and cost efficiency more than frontier-level reasoning.
Pricing at a Glance
| Model | Input (per 1M tokens) | Cached Input | Output (per 1M tokens) | Best for |
|---|---|---|---|---|
| Sol | $5.00 | $0.50 | $30.00 | Complex agentic / research / security work |
| Terra | $2.50 | $0.25 | $15.00 | Production workloads, GPT-5.5 replacement |
| Luna | $1.00 | $0.10 | $6.00 | High-volume classification, extraction, drafting |
The Access Situation
The limited preview is not open to the public. OpenAI shared the models and release plans with the US government before the announcement, as part of what the company describes as a voluntary frontier-model framework. General availability is expected "in the coming weeks," with no specific date confirmed.
OpenAI's statement on the government process: "We don't believe this kind of government access process should become the long-term default." The company is framing this as a one-time step, not a new norm, though the community response has been skeptical.
The @theo thread on X, which drew over 3,600 likes, called this the start of "a dark era in AI model development and access." The criticism centers on the gap between organizations that can access limited previews and those that can't - a concern that's common whenever frontier models ship under restricted access before general availability.
What We Still Don't Know
Because this is a limited preview, several key specs remain unconfirmed:
- Context window size: not disclosed for any of the three tiers
- Latency numbers: no hard figures published yet
- Full benchmark suite: OpenAI has promised more evaluations at general availability
- ChatGPT consumer access: unclear when Sol, Terra, and Luna roll into the standard ChatGPT interface and which plans will get which tiers
- Rate limits: not disclosed for the preview period
- Exact general availability date: "coming weeks" only
We'll update this post when that information becomes public.
The Practical Summary
If you're currently using GPT-5.5 via API for production work, Terra is the obvious migration target, offering the same performance class at half the price. If you're running an agentic pipeline that needs deep reasoning or you work in security research, Sol is what OpenAI built for you. Luna makes sense when you need to process volume cheaply and the task is well-structured enough that frontier reasoning isn't the bottleneck.
The access gate is real but time-limited. For most teams, the strategic move right now is watching the benchmark disclosure that OpenAI plans to release at general launch; that's when the full picture comes into focus.
If you're evaluating which tier to build on for AI coding tools, Cursor and GitHub Copilot are the two most widely-used tools in this space — both draw on OpenAI models. See our Cursor vs GitHub Copilot comparison for a side-by-side breakdown of how they perform in practice.
FAQ
What is GPT-5.6? GPT-5.6 is a new model generation from OpenAI, announced on June 26, 2026, and released initially as a limited preview to approximately 20 organizations. It introduces three capability tiers (Sol, Terra, and Luna), each priced and optimized differently.
What is the difference between GPT-5.6 Sol, Terra, and Luna? Sol is the frontier tier for complex, agentic, and security-sensitive work ($5/$30 per million tokens). Terra is the balanced production tier with GPT-5.5-level performance at half the cost ($2.50/$15). Luna is the high-volume tier for fast, structured tasks at the lowest price ($1/$6).
How much does GPT-5.6 cost? Sol costs $5 per million input tokens and $30 per million output tokens. Terra is $2.50 input / $15 output. Luna is $1 input / $6 output. All three support a 90% discount on cached input reads.
When will GPT-5.6 be publicly available? OpenAI has said general availability is expected "in the coming weeks" from the June 26, 2026 announcement. No specific date has been confirmed.
Why is GPT-5.6 only available to select partners? OpenAI shared the models with the US government before release as part of a voluntary frontier-model framework. The company has stated this access process is not intended to be the long-term default. The limited preview is expected to expand to general availability within weeks.
What is GPT-5.6 Sol ultra mode? Ultra mode is a new reasoning setting in Sol that deploys parallel subagents to work on different parts of a complex task simultaneously. It trades latency for higher accuracy and is how Sol achieved its top Terminal-Bench 2.1 score of 91.91%.