📅 2026-07-10 · 11 min read
GPT-5.6 Launch Guide: Sol vs Terra vs Luna, Pricing, Benchmarks, and Who Should Upgrade
GPT-5.6 is not one model but a three-tier family: Sol for maximum capability, Terra for balanced everyday work, and Luna for speed and cost. This guide turns OpenAI's launch data into a practical selection framework.
Launch snapshot
GPT-5.6 family at a glance
Official list pricing per 1M tokens. Capability tiers serve different workload and cost profiles.
📑 Table of Contents
What launched
OpenAI moved the GPT-5.6 family from limited preview to general availability on July 9, 2026. The family includes Sol, the flagship model; Terra, a balanced lower-cost tier for everyday work; and Luna, the fastest and least expensive tier.
The naming matters because the number identifies the model generation while Sol, Terra, and Luna are durable capability tiers. That means the practical decision is no longer simply whether to use GPT-5.6, but which tier best matches task difficulty, latency needs, and budget.
The core selection rule
Start with Terra for balanced everyday work, move to Sol when task failure is expensive or the workflow is genuinely difficult, and use Luna when scale, speed, and cost matter more than maximum reasoning depth.
Availability and pricing
OpenAI says GPT-5.6 is available across ChatGPT, Codex, and the OpenAI API, with rollout continuing gradually toward full availability. In the API, all three tiers are available, and the Responses API adds Programmatic Tool Calling plus a multi-agent beta for concurrent subagents.
| Model | Input / 1M tokens | Output / 1M tokens | Best first use |
|---|---|---|---|
| GPT-5.6 Sol | $5 | $30 | Hard coding, long-running agent work, complex professional tasks |
| GPT-5.6 Terra | $2.50 | $15 | Balanced daily work, production agents, cost-sensitive reasoning |
| GPT-5.6 Luna | $1 | $6 | High-volume, lower-latency, economical workloads |
OpenAI also changed prompt caching economics for GPT-5.6 and later models: cache writes are billed at 1.25× the uncached input rate, while cache reads retain a 90% cached-input discount. Check current API docs before hard-coding budget assumptions.
Benchmark highlights
The launch data shows a broad pattern: Sol leads or approaches the frontier on several agentic, coding, browsing, computer-use, cybersecurity, and science evaluations, while Terra and Luna often trade some peak score for better cost and speed.
Official launch benchmarks
Selected GPT-5.6 scores
Scores below come from OpenAI's launch page. Compare within the same benchmark and evaluation setup.
| Benchmark | Sol | Terra | Luna | GPT-5.5 |
|---|---|---|---|---|
| Coding Agent Index v1.1 | 80.0 | 77.4 | 74.6 | 76.4 |
| DeepSWE v1.1 | 72.7% | 69.6% | 67.2% | 67.0% |
| Terminal-Bench 2.1 | 88.8% | 87.4% | 84.7% | 85.6% |
| BrowseComp | 90.4% | 87.5% | 83.3% | 84.4% |
| OSWorld 2.0 | 62.6% | 50.2% | 45.6% | 47.5% |
| ExploitBench | 73.5% | 52.9% | 33.2% | 47.9% |
One important caveat: no single table proves universal superiority. Some launch tables include competitor models that lead individual benchmarks, and different harnesses, reasoning settings, latency budgets, and token budgets can change the ordering.
Coding and agent work
OpenAI positions Sol as its strongest coding model so far. On the Artificial Analysis Coding Agent Index v1.1, Sol reaches 80, while Terra reaches 77.4 and Luna 74.6. On Terminal-Bench 2.1, Sol scores 88.8%, Terra 87.4%, and Luna 84.7%.
The more important product change may be workflow rather than raw score. GPT-5.6 can use Programmatic Tool Calling to write and run lightweight programs that coordinate tools, process intermediate results, and decide next actions. OpenAI also introduces max reasoning and an ultra mode that coordinates multiple agents in parallel for demanding tasks.
Use Sol when
A failed implementation, long debugging loop, or complex agent workflow costs more than the model premium.
Use Terra when
You need strong coding-agent performance but want tighter control of production cost.
Use Luna when
The task is frequent, parallelizable, and benefits more from speed and volume than maximum depth.
Knowledge work and computer use
The launch is also aimed at work beyond coding. OpenAI reports Sol at 90.4% on BrowseComp and 62.6% on OSWorld 2.0. The company says GPT-5.6 is better at turning messy context from documents and connected work systems into polished deliverables, including presentations, documents, spreadsheets, and interactive artifacts.
For buyers, this matters because the upgrade case may come from fewer revisions and fewer tool round trips rather than a single visible benchmark win. Teams should measure accepted output, correction time, task completion rate, and total cost—not just model price per token.
Cybersecurity and science
GPT-5.6 shows large gains in several security and science evaluations. On ExploitBench, Sol scores 73.5% versus GPT-5.5 at 47.9%. On ExploitGym, Sol reaches 33.7% under the longer time budget compared with GPT-5.5 at 15.1% under the reported shorter baseline. The launch page also reports gains across GeneBench Pro, LifeSciBench, and internal medicinal chemistry evaluation.
These stronger capabilities also raise governance requirements. OpenAI's launch materials emphasize layered safeguards, monitoring, access calibrated to trust and risk, and an updated system card. High-capability cyber use should be assessed with explicit authorization, scope, logging, and review.
Who should upgrade from GPT-5.5?
| Workflow | First model to test | Why |
|---|---|---|
| Hard repo-scale coding or terminal agent work | Sol | Strongest coding and terminal scores in the GPT-5.6 family. |
| General professional work and production agents | Terra | Balanced capability and lower cost; strong first upgrade candidate. |
| High-volume summarization, routing, extraction, or parallel tasks | Luna | Lowest list price and fastest tier. |
| Stable GPT-5.5 workflow with no quality bottleneck | Run a controlled pilot first | Do not migrate only because a new model exists; compare end-to-end cost and accepted output. |
For teams already using GPT-5.5 successfully, the right migration test is representative workload A/B testing. Measure completion rate, retries, output acceptance, latency, token usage, and review burden across the same task set.
FAQ
Is GPT-5.6 one model?
No. It is a family with Sol, Terra, and Luna capability tiers.
How much does GPT-5.6 cost in the API?
OpenAI lists Sol at $5 input and $30 output per 1M tokens, Terra at $2.50/$15, and Luna at $1/$6.
Where is GPT-5.6 available?
OpenAI says the family is rolling out across ChatGPT, Codex, and the API, with plan and feature availability varying by surface.
Should everyone move from GPT-5.5 immediately?
No. Test the tier that matches your workload and compare full task economics, not only token price or headline benchmark scores.
Primary source: OpenAI's official GPT-5.6 launch page. Benchmark and pricing values in this guide are based on that release material.