OpenAI ships GPT-5.6 with three tiers and Copilot deal

TL;DR:

  • OpenAI has made GPT-5.6 generally available across ChatGPT, Codex and its API, split into three tiers: flagship Sol, mid-range Terra and low-cost Luna.
  • API pricing runs from $5/$30 per million input/output tokens for Sol down to $1/$6 for Luna, with the company pitching “performance per dollar” over raw capability.
  • The model became the preferred engine in Microsoft 365 Copilot the same day, putting it inside Word, Excel, PowerPoint and Chat.

OpenAI has released GPT-5.6, the model whose launch Washington paused last month on security grounds, and the framing has shifted from raw capability to cost. The company is selling three durable tiers — Sol, Terra and Luna — and leaning hard on the argument that most work no longer needs its most expensive model.

Priced for the finance team

The economics are the story. Sol costs $5 per million input tokens and $30 per million output; Terra is priced at $2.50/$15, and Luna at $1/$6. OpenAI claims Terra and Luna match or beat rivals at a fraction of the cost, and one analyst quoted by Reuters found the smallest model could complete a task about as well as the largest at roughly one-fifth of the price. A new ultra setting coordinates several agents in parallel for the hardest jobs, trading tokens for speed. For buyers who spent last year worrying about runaway inference bills — a concern fresh KPMG data puts at the centre of stalled AI projects — a cheaper default tier matters more than another benchmark record.

Distribution arrived the same day. Microsoft named GPT-5.6 the preferred model in Microsoft 365 Copilot, wiring it into Word, Excel, PowerPoint, Copilot Chat and its Cowork agent. That channel reaches millions of UK enterprise desktops without users choosing a model at all — the “which model is actually running in your organisation” question is now answered by default settings rather than procurement.

Looking forward

For UK firms, GPT-5.6 sharpens a decision that is becoming financial rather than technical: which tier, at what cost, for which task. The performance-per-dollar pitch also intensifies pressure on rivals as cheaper Chinese models close the gap. Expect procurement teams to start treating model choice like any other cloud line item — matching Luna, Terra or Sol to workloads rather than defaulting to the flagship.

Share this article