GPT-5.6 Series Launch: Three-Tier Models, Price War, and the Multi-Agent Era
OpenAI officially launched the GPT-5.6 series featuring three tiers—flagship Sol, balanced Terra, and cost-optimized Luna—alongside max deep reasoning and ultra multi-agent modes, reshaping the pricing and capability landscape for large language models.
GPT-5.6's launch is unlike any before it—not a single model, but a family.
On July 9, 2026, after nearly two weeks of limited preview, OpenAI officially rolled out the GPT-5.6 series to the world. This time, OpenAI ditched the "one model to rule them all" playbook and came to the table with three cards: the flagship Sol, the balanced Terra, and the lightweight Luna.
Three Tiers, Full Coverage
GPT-5.6's naming logic is refreshingly simple: the number (5.6) denotes the generation, while Sol / Terra / Luna represent capability tiers that will persist and iterate at their own cadence.
| Model | Positioning | Input ($/1M tokens) | Output ($/1M tokens) |
|---|---|---|---|
| Sol | Flagship — coding, research, security SOTA | 5.00 | 30.00 |
| Terra | Balanced — GPT-5.5 performance at lower cost | 2.50 | 15.00 |
| Luna | Cost-optimized — high-throughput, low-latency | 1.00 | 6.00 |
Sol is the undisputed champion, built for the hardest problems in coding, cybersecurity, and scientific research. Terra's positioning is particularly compelling: it matches the previous flagship GPT-5.5 in performance while costing half as much. And Luna, at $1 per million input tokens, lowers the barrier to entry dramatically.
The genius of this three-tier strategy is that OpenAI no longer expects every user to pay for flagship capability. Terra for daily work, Luna for batch processing, and Sol only when you need to tackle the truly hard stuff.
A Real Price War
Stack GPT-5.6's pricing against competitors and this isn't just a "price cut"—it's a repricing of the entire category.
Terra delivers GPT-5.5-grade performance at half the cost. Luna is even more impressive: at roughly one-sixteenth the cost, it still outscores Anthropic's Claude Fable 5 on key benchmarks. OpenAI's messaging is clear: every token now goes further.
For developers, this means the same budget buys significantly more work. Classification, content extraction, and large-scale summarization—tasks that once required careful budgeting—can now be handled comfortably by Luna.
Max and Ultra: Two New Modes
GPT-5.6 introduces two new reasoning modes—easy to overlook, but genuinely important.
Max mode gives the model extended reasoning time to explore alternatives, self-check, and revise before delivering a final answer. It's not just "think harder"—it's a qualitatively deeper internal deliberation process.
Ultra mode goes even further: it coordinates four AI agents by default to work on tasks in parallel, expandable to sixteen via the Responses API's multi-agent configuration. This marks the first time OpenAI has baked native multi-agent orchestration into a flagship model—essentially spinning up a miniature AI team within a single API call.
Performance: New SOTA in Coding
On the Artificial Analysis Coding Agent Index v1.1, GPT-5.6 Sol (max reasoning) scored 80 points—a public SOTA. On Terminal-Bench 2.1 for command-line workflows, it achieved 88.8% in standard mode and 91.9% in ultra mode.
What's even more telling is the efficiency. On the Agents' Last Exam benchmark, Sol scored 53.6, beating Claude Fable 5 by 13.1 percentage points, while costing only a quarter as much.
More Than a Model Launch
Alongside GPT-5.6 came ChatGPT Work—OpenAI's answer to Claude Cowork and WorkBuddy. Codex was also merged into ChatGPT, ending the artificial split between code and conversation.
This is a clear signal: GPT-5.6 isn't the finish line—it's a key move in OpenAI's strategy to build a moat around "model × tooling × pricing." For developers, the options have multiplied, and so has the cognitive load of choosing among them. The silver lining? The naming scheme finally makes sense—just learn Sol / Terra / Luna, then pick based on what you need.