GPT-5.6 Luna: 80% Price Cut Puts Budget AI Ahead of Gemini Flash

The 80% Price Cut in Detail

On 30 July 2026 — just three weeks after the launch of GPT-5.6 on 9 July — OpenAI cut pricing for its entry-level model Luna by 80%: input tokens dropped from $1.00 to $0.20 per million, and output tokens from $6.00 to $1.20 per million. The mid-tier model Terra also became cheaper, though more moderately, falling 20% to $2.00/$12.00 per million tokens (previously $2.50/$15.00). The flagship model Sol remains unchanged at $5.00/$30.00.

Why a Price War Changes the Game

This price cut is more than a tactical discount. It shifts the question procurement and IT leaders ask from 'which model is the most capable?' to 'which model delivers the most intelligence per dollar spent?' Organisations that haven't yet built this calculation into their AI strategy are leaving budget on the table.

Luna Head-to-Head: Gemini, DeepSeek and OpenAI's Own Family

• GPT-5.6 Luna: $0.20 / $1.20 per million tokens, Intelligence Index 51.2
• Gemini 3.5 Flash-Lite: $0.25 / $1.50 per million tokens — more expensive than Luna
• Gemini 3.6 Flash: $1.50 / $7.50 per million tokens, Intelligence Index 50.2 — weaker than Luna despite a far higher price
• DeepSeek V4 Flash: $0.14 / $0.28 per million tokens — cheaper on raw token price, but with a noticeable performance gap
• GPT-5.6 Terra: $2.00 / $12.00 per million tokens, Intelligence Index 55
• GPT-5.6 Sol: $5.00 / $30.00 per million tokens, Intelligence Index 58.9
• Fable 5 remains the overall top performer in the Artificial Analysis Intelligence Index at 59.9

The result, according to the Artificial Analysis Intelligence Index, is striking: Luna now scores ahead of Gemini 3.6 Flash — at a fraction of the price. Only DeepSeek V4 Flash undercuts Luna on raw token pricing, but it falls behind on intelligence. For tasks where reliability matters, the pricier Luna can ultimately be cheaper than the less expensive DeepSeek model.

What This Means for Swiss SMEs

For Swiss SMEs, the entry barrier for productive AI use has dropped again, and noticeably so. Total cost of ownership for standard tasks — text processing, classification, simple automation — has fallen to a level that would have been unthinkable just months ago.

Model-Agnostic Architecture Becomes Strategic

When the price-performance balance between providers shifts within weeks, the ability to switch models flexibly by task becomes a competitive advantage in its own right. Companies that lock their AI architecture to a single provider or model risk chronically overpaying or leaving performance on the table.

What This Means for Choosing Your AI Partner

As an outsourced AI division, this is exactly where we operate: we build architectures that can dynamically choose between models such as Luna, Terra, Sol, Gemini or DeepSeek — matched to task, budget and quality requirements. That way, you automatically benefit from every price cut and every performance leap without rebuilding your infrastructure each time.

Looking Ahead

The price war between OpenAI, Google and Chinese providers such as DeepSeek is likely to continue. For decision-makers, the takeaway is clear: investing today in a flexible, model-agnostic AI strategy secures the best terms tomorrow — regardless of which provider happens to be ahead at any given moment.