Quick answer. GPT-5.6 is real and shipping. OpenAI previewed it on June 26, 2026 as a three-tier family — Sol (flagship), Terra (balanced), and Luna (fast and cheap) — starting with a limited partner preview via the API and Codex, with general availability rolling out in July 2026. For most production work, OpenAI positions Terra as GPT-5.5-level quality at roughly half the cost; reach for Sol (or Sol Ultra) on the hardest agentic and coding tasks.
🆕 Head-to-head: See how GPT-5.6 compares with Anthropic's flagship in GPT-5.6 vs Claude Fable 5 — a neutral breakdown of price, benchmarks, and coding.
This post used to track GPT-5.6 while it was still a rumor. That phase is over. On June 26, 2026, OpenAI previewed GPT-5.6 for real — and it did not ship the way earlier point releases did. Instead of one frontier model with adjustable effort, GPT-5.6 arrives as three named tiers, each tuned to a different point on the cost, speed, and capability curve. General availability is rolling out in July 2026.
Here is what is confirmed, what to actually run, and where the numbers still need independent verification.
What is the GPT-5.6 release date?
The GPT-5.6 release date was June 26, 2026, when OpenAI previewed it — initially as a limited-access release to a small set of partner organizations through the API and OpenAI Codex. At the U.S. government's request, OpenAI started with a restricted preview rather than an open launch, and has said it does not expect that access process to become the long-term default. Broader general availability is rolling out in July 2026. So the correct framing is no longer "unreleased" — it is "released and opening up."
What are the GPT-5.6 tiers (Sol, Terra, Luna)?
The biggest change in GPT-5.6 is the naming and packaging. The number (5.6) marks the generation; Sol, Terra, and Luna are durable model tiers you pick between, not effort toggles on a single model.
| Tier | Built for (per OpenAI) | Price (input / output per 1M) |
|---|---|---|
| Sol (flagship) | Frontier coding, agents, biology, cybersecurity | $5.00 / $30.00 |
| Terra (balanced) | Everyday production — OpenAI targets GPT-5.5 quality at ~2× lower cost | $2.50 / $15.00 |
| Luna (fast) | High-volume, latency-sensitive work; chatbots, classification | $1.00 / $6.00 |
Sol also offers higher reasoning-effort settings, and Sol Ultra is its top configuration for the hardest reasoning and agentic tasks — at the same price as base Sol ($5 / $30). Prompt caching is supported across all tiers, which matters a lot for cost on repetitive agent loops.
GPT-5.6 benchmarks: how much better is it?
On Terminal-Bench 2.1 (agentic coding), OpenAI reports:
- Sol Ultra: 91.9%
- Sol (base): 88.8%
- Claude Mythos 5: 88.0%
- GPT-5.5: 88.0%
Read that honestly. Sol Ultra's reported ~3.9-point lead is meaningful but not a landslide, and base Sol's reported edge over GPT-5.5 (0.8 points) is narrow. OpenAI's system card also reports gains on SecureBio biology evaluations (exact scores unpublished) and says the models stay below its "Cyber Critical" threshold — i.e., not capable of autonomous, full-chain cyberattacks in OpenAI's own tests. These are OpenAI's own numbers; treat them as directional until independent, third-party testing lands after general availability.
GPT-5.6 vs GPT-5.5: what actually changed?
Three things matter for a working developer:
- Tiering. You now choose Sol / Terra / Luna up front instead of tuning one model. Terra is the headline value: OpenAI positions it as GPT-5.5-class quality at roughly half the price.
- Context window. GPT-5.6 reportedly ships a 1.5M-token context window, up from the 1M GPT-5.5 shipped with.
- Capability. Modest reported gains on the hardest coding, agentic, biology, and cybersecurity evaluations — incremental rather than a generational jump.
We go deeper on the upgrade decision in GPT-5.6 vs GPT-5.5: what changed and should you upgrade, and on the coding head-to-head in GPT-5.6 vs Claude Opus 4.8. For the full tier breakdown, see the GPT-5.6 Sol, Terra and Luna guide.
What should you run right now?
- Most production workloads → Terra. OpenAI positions it as GPT-5.5 quality at ~2× lower cost, so it's the first tier to benchmark if cost reduction is the goal.
- Hardest agentic coding / research → Sol or Sol Ultra. Worth the premium only when the task genuinely needs the top of the range.
- High-volume, latency-sensitive (chatbots, classification) → Luna. Cheapest and fastest.
- Not yet on the preview allowlist? Stay on GPT-5.5 (or GPT-5.5 Instant for low-latency work) until GA reaches your account, then benchmark Terra against your current model with your own prompts before switching.
For where GPT-5.5 still fits and how the family evolved, see our GPT-5.5 complete developer guide. If you're weighing OpenAI against Anthropic, the Claude Opus complete guide is the companion read.
How will you know when GPT-5.6 is fully available?
Watch primary sources, not screenshots: OpenAI's official introducing-GPT-5.6 post and model card on openai.com, and the moment the gpt-5.6 tiers (Sol / Terra / Luna) appear as callable models in the OpenAI API model list for your account. When those are live for you, GPT-5.6 is fully available — regardless of what any third-party article claims.
FAQ
When was GPT-5.6 released?
OpenAI previewed GPT-5.6 on June 26, 2026 as a limited partner release via the API and Codex, with general availability rolling out in July 2026.
What are Sol, Terra, and Luna?
They are the three durable tiers of GPT-5.6. Sol is the flagship ($5/$30 per million tokens), Terra is the balanced everyday model ($2.50/$15, which OpenAI positions as GPT-5.5 quality at about half the cost), and Luna is the fast, cheap option for high-volume work ($1/$6). Sol Ultra is Sol's top high-effort configuration at the same price.
How much does GPT-5.6 cost?
Per million tokens: Sol and Sol Ultra $5 input / $30 output, Terra $2.50 / $15, and Luna $1 / $6. Prompt caching is available on all tiers.
Is GPT-5.6 better than GPT-5.5?
Modestly, on OpenAI's own numbers. On Terminal-Bench 2.1, OpenAI reports Sol Ultra scored 91.9% and base Sol 88.8% versus GPT-5.5's 88.0%. The bigger practical wins are the Terra tier's lower cost and the larger 1.5M-token context window.
What context window does GPT-5.6 have?
GPT-5.6 reportedly ships a 1.5 million token context window, up from the 1 million tokens GPT-5.5 shipped with.
Standardizing your team on frontier models and agentic coding tools? Codersera helps you hire vetted remote developers who already build with GPT, Claude, and Codex in the loop.