AI Prices Are Collapsing: What the 2026 Price War Means for Swiss SMEs

On July 30, 2026, OpenAI cut prices for GPT-5.6 Luna by 80% — from $1.00 to $0.20 per million input tokens, with output pricing falling from $6.00 to $1.20. Terra, the next tier up, became 20% cheaper. What stands out is not just the size of the cut but the timing: Luna had been on the market for only a short time. A price cut of this magnitude so soon after launch is not a gesture of generosity — it is a sign of competitive pressure.

What exactly happened on July 30, 2026?

OpenAI restructured pricing across three model tiers. Luna, the entry-level model, was slashed dramatically. Terra, the mid-tier model, more moderately. Sol, the flagship, stayed unchanged at $5 and $30 per million tokens. This three-way split is itself a signal: OpenAI is protecting margin at the premium tier while competing on volume and market share at the entry and mid tiers — precisely where competitive pressure from open models is strongest.

1. Luna: input $1.00 → $0.20 per million tokens (−80%), output $6.00 → $1.20 (−80%)
2. Terra: input $2.50 → $2.00 per million tokens (−20%), output $15.00 → $12.00 (−20%)
3. Sol: unchanged at $5.00 / $30.00 per million tokens

A detailed breakdown of this pricing round and how it compares to Google's entry-level model is available in our analysis of GPT-5.6 Luna's 80% price cut versus Gemini Flash.

The real driver: open-source models are pulling the price/intelligence curve down

This price cut is not an isolated OpenAI decision — it is a forced response. According to MLQ.ai, Luna's new input price now actually undercuts DeepSeek V4 Pro ($0.435 per million tokens), though DeepSeek remains cheaper on output. Add to that Kimi K3, an open model with 2.8 trillion parameters released under an Apache 2.0 license, competing with no licensing fees and full customizability. These open models are not competing primarily on features — they are competing on cost per unit of intelligence, and that is reshaping the entire industry.

CNBC, citing OpenRouter data, reports that Chinese models already account for 46% of US enterprise token usage. This is no longer a niche phenomenon — it is a structural market share that puts direct pressure on closed providers.

The long-term trend: from $60 to six cents

Writing in Forbes, Peter Cohan traced the trajectory of token costs since 2021: comparable capability cost roughly $60 back then; by July 2026, the price had fallen to $0.06. That is a thousandfold reduction in five years. This curve is not linear — it is accelerating. Every new open model generation shifts the benchmark that closed providers are measured against.

Anthropic and Google are moving in the same direction. Claude Opus 5 is priced at $5 and $25 per million tokens, roughly 6% below OpenAI's Sol. Google's Gemini Flash sits at $1.50 and $7.50. For a systematic look at which model fits which budget and use case, our guide on intelligence per dollar in 2026 offers a practical framework.

Why OpenAI is under pressure despite its market lead

OpenAI's own Q1 2026 shareholder letter reveals the scale of the problem: average monthly growth (AGM) came in at 33%, against an internal target of 46%. That gap between ambition and reality explains why a company with an enormous market position would slash the price of a model launched just three weeks earlier so drastically. Growth, not margin, is currently the decisive metric — and growth is fastest to buy through price.

What this means concretely for Swiss SMEs

For Swiss decision-makers, this price collapse carries three direct consequences. First, the classic budget argument against AI adoption is losing substance. Enterprise DNA analyses show that tasks whose ROI calculation was negative just three weeks ago are already profitable at Luna's new prices. Second, access to frontier capability is no longer a differentiator — it has become a market signal, not a secret. The real competitive edge now lies in who most efficiently organizes cost per actual business outcome. Third, model agility is becoming a strategic core competency. Companies that lock themselves into a single provider systematically miss the next price cut or the next capability leap.

This shift requires a different mindset than the one that prevailed two years ago. Instead of a one-time 'which model fits us' decision, companies need an ongoing capability to switch between models, compare providers, and build architectures where a model swap does not require a rebuild.

Vendor diversification instead of model loyalty

The obvious reaction for many companies so far has been to pick one provider and stick with it — for simplicity, integration cost, or plain inertia. The current price collapse makes that strategy risky. A company relying exclusively on one closed model today may be paying several times more in six months than a multi-model architecture would require. The reverse is also true: switching too early and too completely to a single open-source model risks dependency on less mature infrastructure.

The pragmatic middle path is deliberate vendor diversification: different models for different task classes, regular reassessment of the price-performance landscape, and a technical architecture that turns switching providers into a configuration decision rather than a project. This capability — not access to any particular model — is becoming the real competitive advantage.

Why AI outsourcing is the right answer to this market dynamic

This is precisely where the case for AI outsourcing as an operating model becomes clear. An internal AI team that has to evaluate every price change, every new model and every licensing shift on its own will structurally fall behind at this pace. An external partner who continuously tracks model performance and pricing across multiple providers can select the most economical combination at any given time — without the Swiss SME itself having to become a full-time observer of global AI pricing. Model agility stops being an extra task and becomes a purchased standard.

Bottom line: price is no longer the hurdle — organization is

The message of July 30, 2026 is unambiguous: powerful AI has become financially accessible to virtually every Swiss SME. The question is no longer whether a company can afford AI, but whether it is organizationally capable of continuously capturing the best available price-performance ratio. Companies that do not want to build this agility in-house will find the structural advantage the market now demands in a specialized outsourcing partner.