After weeks of waiting — and an unusual government-mandated hold — OpenAI's GPT-5.6 is live today. The new model family includes three variants: Sol (flagship), Terra (balanced), and Luna (fast and affordable). Here's what you need to know.
Why Was GPT-5.6 Delayed?
On June 26, OpenAI announced it was restricting the initial rollout at the US government's request. Only around 20 pre-approved partners could access it while the Commerce Department's Center for AI Standards and Innovation ran a security review.
The concern? GPT-5.6 showed significant jumps in cybersecurity and biological research capabilities — exactly the kind of dual-use power that Washington worries could be exploited by adversaries. OpenAI complied but made clear this kind of delay "shouldn't be the norm."
The clearance came through. GPT-5.6 officially launches today, July 9.
| Model | Tier | Best For | Price (per 1M tokens) | Standout Feature |
|---|---|---|---|---|
| GPT-5.6 Sol | Flagship | Agentic workflows, research, coding | $5 in / $30 out | Ultra Mode (multi-agent, 91.9% benchmark) |
| GPT-5.6 Terra | Balanced | Everyday dev & business tasks | $2.50 in / $15 out | GPT-5.5 performance at 2× lower cost |
| GPT-5.6 Luna | Fast & Affordable | High-volume, cost-sensitive apps | $1 in / $6 out | Fastest tier; "High" safety rating at budget price |
The Three Models Explained
GPT-5.6 Sol — Flagship
Sol is OpenAI's most capable model yet. It tops Terminal-Bench 2.1 at 88.8% in standard mode, jumping to 91.9% in Ultra Mode — ahead of Claude Fable 5 (83.4%), Claude Mythos 5 (84.3%), and Gemini 3.1 Pro Preview (70.7%).
The standout feature is Ultra Mode: Sol decomposes a complex task internally and spawns parallel subagents, each handling a piece before synthesizing the result. It's the first GPT model where this multi-agent orchestration is native, not a wrapper. OpenAI is also running Sol on Cerebras hardware at up to 750 tokens per second.
Price: $5 input / $30 output per million tokens.
GPT-5.6 Terra — Balanced
Terra delivers GPT-5.5-level performance at roughly half the cost. For most developers and businesses, this will be the practical daily driver — frontier-adjacent intelligence without Sol's price tag.
Price: $2.50 input / $15 output per million tokens.
GPT-5.6 Luna — Fast & Affordable
Luna is built for speed and cost. Despite being the budget tier, it still carries OpenAI's "High" safety classification for cybersecurity and biological risk — a first for any budget-tier GPT model, and a signal of how capable even the cheapest option is.
Price: $1 input / $6 output per million tokens.
Benchmark Snapshot
| Model | Terminal-Bench 2.1 |
|---|---|
| GPT-5.6 Sol Ultra | 91.9% |
| GPT-5.6 Sol | 88.8% |
| GPT-5.5 | 88.0% |
| Claude Mythos 5 | 84.3% |
| Claude Fable 5 | 83.4% |
| Claude Opus 4.8 | 78.9% |
| Gemini 3.1 Pro Preview | 70.7% |
| Model | Score | vs GPT-5.5 |
|---|---|---|
| GPT-5.6 Sol Ultra | 91.9% | +3.9% |
| GPT-5.6 Sol | 88.8% | +0.8% |
| GPT-5.5 (previous flagship) | 88.0% | — |
| Claude Mythos 5 | 84.3% | −3.7% |
| Claude Fable 5 | 83.4% | −4.6% |
| Claude Opus 4.8 | 78.9% | −9.1% |
| Gemini 3.1 Pro Preview | 70.7% | −17.3% |
The Safety Question
All three GPT-5.6 models carry OpenAI's "High" Preparedness Framework classification for both cybersecurity and biological/chemical risk. This is the first GPT family where even the cheapest model triggered the highest non-Critical safety tier. The government review process included additional testing before today's green light.
Availability
GPT-5.6 Sol, Terra, and Luna are available today in global preview via ChatGPT and the API. OpenAI confirmed broader general availability will expand in the coming weeks. ChatGPT Pro users get Sol-tier access.
My Take
The government delay is the real story here, and I think it's one people are sleeping on. We've quietly crossed a threshold where a commercial AI model is now powerful enough that Washington wants a look before the public does. That's new territory — and it's not going back.
OpenAI complied but made clear they weren't happy about it. That tension matters. The more capable these models get, the more we're going to see this push-and-pull between labs moving fast and governments trying to catch up. GPT-5.6 won't be the last model held back for a security review.
As for the models themselves: Terra is the obvious daily driver for most people and teams. Sol in Ultra Mode is genuinely impressive on the benchmarks, but unless you're running heavy multi-step agentic pipelines, the difference won't feel dramatic in practice. Luna is the sleeper pick — more capable than any budget model has a right to be, and the pricing makes it easy to run at scale without worrying about costs.
The fact that even Luna carries OpenAI's highest non-Critical safety rating says everything about where AI capability is right now. We're not in GPT-3 territory anymore.
Frequently Asked Questions
What is GPT-5.6?
GPT-5.6 is OpenAI's latest model generation, available in three tiers: Sol (flagship), Terra (balanced), and Luna (affordable/fast).
Why was GPT-5.6 delayed?
The US government requested early access for a security review before public launch, citing concerns about the model's advanced cybersecurity and coding capabilities.
What is Ultra Mode?
Ultra Mode is a built-in multi-agent system in Sol where the model decomposes complex tasks and runs parallel subagent processes simultaneously before synthesizing results.
How much does GPT-5.6 cost?
Sol is $5/$30 per million tokens, Terra is $2.50/$15, and Luna is $1/$6 (input/output).
Which GPT-5.6 model should I use?
Terra for most everyday and developer tasks. Sol for heavy agentic or research workflows where performance is critical. Luna for high-volume, cost-sensitive applications.
