TL;DR

OpenAI reduced prices on its lower and middle GPT-5.6 tiers on Thursday. Luna, the smallest, falls 80%; Terra, the mid-tier, falls 20%. Sol, the flagship, is unchanged. Per million tokens, Luna’s input price falls from $1 to 20 cents and Terra’s from $2.50 to $2, while output generation drops from $6 to $1.20 and from $15 to $12 respectively.

The company attributes the cuts to efficiency gains in GPT-5.6, including the model’s own contribution to optimising its code. The more plausible reading is that buyers forced it. Enterprise customers have spent this year interrogating AI invoices, and open-weight Chinese models such as Z.ai’s GLM-5.2 now land close enough on performance at materially lower cost to make the comparison uncomfortable.

Anthropic is the other party under pressure. Claude Sonnet 4.6, its mid-tier, charges $3 for a million input tokens and $15 for a million output — now above Terra on both counts. Anthropic’s models hold a strong position in enterprise and developer work, but they sit at the expensive end, and that end just got further from the middle.

There is a catch worth carrying into any UK procurement conversation. Token prices have been falling steadily for a year, yet the cost of finishing a given task has been going up. Reasoning models consume far more tokens per task than their predecessors, and the industry’s drift from flat subscriptions towards usage-based billing means the bill scales with how hard the model decides to think. Cheaper units, more units.

For UK firms, the practical effect is that a headline price cut is not a budget cut. It changes which tier a workload should sit on — OpenAI’s own argument is that Luna and Terra can now handle jobs that recently needed the flagship — but it does nothing to make monthly spend predictable. That remains the harder problem, and one no vendor has solved.

Looking Forward

Analysts have flagged that price cuts may lift usage while straining margins at both OpenAI and Anthropic, awkward timing given the initial public offerings both are widely expected to pursue. Expect Anthropic to respond on price or on tier structure. For buyers, the sensible move is to re-benchmark existing workloads against the cheaper tiers rather than assume the saving arrives automatically.