GPT-5.6 Sol’s 50% Gateway Discount Puts Pressure on Anthropic’s AI Pricing Lead

Why the discount matters in the OpenAI–Anthropic contest

The promotion attacks one of Anthropic’s clearest advantages: its ability to convert a smaller share of token traffic into a much larger share of spending.

Vercel’s July AI Gateway index reported that Anthropic captured 61% of gateway spending from 32% of token volume, while open-weight models handled 29% of tokens on less than 4% of spend. In a later Vercel update, Anthropic was reported to hold 65% of spending from 30% of token volume, with an average price per token 4.4 times that of other labs.

Those figures point to two different contests:

• Token volume: how much inference traffic a provider handles.
• Monetization: how much spending that traffic generates.

OpenAI can improve its position in the first contest by making Sol cheaper and easier to access. But to close Anthropic’s monetization gap, it must also persuade customers that Sol is reliable enough for high-value production work, not merely attractive for temporary evaluations or price-sensitive workloads.

The promotion could still be strategically valuable even if it reduces revenue per token in the short term. More usage gives OpenAI additional exposure to real developer workloads, creates opportunities for application teams to switch providers, and makes the model more familiar to developers who might otherwise default to Claude.

OpenAI is competing in a three-way market, not a two-way race

The OpenAI–Anthropic comparison captures only part of the market. OpenRouter usage data shows that Chinese models are taking a large share of token volume.

For the week of July 28 through August 3, Chinese models processed 28.13 trillion of 56.8 trillion total tokens on OpenRouter, while nine of the ten most-used models were Chinese, according to CGTN’s report on the platform’s rankings. Another analysis of monthly OpenRouter usage placed Xiaomi’s MiMo‑V2.5 at 32.8 trillion tokens and DeepSeek V4 Flash at 26.4 trillion, compared with 8.6 trillion for OpenAI’s GPT‑5.6 Luna and 4.8 trillion for Claude Opus among closed models.

That pattern suggests a barbell-shaped market:

1. Low-cost or open-weight models absorb large-scale, price-sensitive, and increasingly agentic workloads.
2. OpenAI uses model improvements, distribution, and targeted discounts to compete for mainstream proprietary demand.
3. Anthropic retains a strong position in production workloads where customers are willing to pay more for model behavior, reliability, and performance.

Vercel’s data reinforces the separation between scale and revenue: open-weight models can handle a substantial share of tokens without capturing a comparable share of spending. A model can therefore “win” usage rankings while remaining a relatively small commercial business compared with a more expensive model used for consequential tasks.

The gateway layer is becoming strategically important

The promotion also highlights the growing importance of model gateways. OpenRouter gives developers a way to compare and switch among models through a common access layer; reporting on the company’s acquisition describes access to hundreds of models and a developer base of roughly 8 million, rather than the larger figure sometimes repeated elsewhere.

Bloomberg reported that Stripe finalized an agreement to acquire OpenRouter for more than $7 billion, citing people familiar with the matter. Other reports note that the final price could still change and that the transaction was not yet publicly confirmed by all parties.

If completed, the deal would underline the strategic value of the layer between AI developers and model providers. Routing, billing, observability, and provider substitution allow developers to make model choices based on price, latency, quality, and availability rather than committing permanently to one laboratory.

That flexibility increases competitive pressure on both OpenAI and Anthropic. It also makes temporary promotions more powerful: a gateway can expose a discounted model to existing applications without requiring every developer to rebuild its integration.

What to watch after September 18

The most important test is what happens when the Sol promotion ends. Three outcomes would have different implications:

• Usage falls sharply: the offer mainly attracted price-sensitive experiments and did not change production defaults.
• Usage remains high but spending does not: OpenAI gains distribution and token volume, while Anthropic retains the premium workload advantage.
• Usage and spending both remain strong: Sol has converted a temporary discount into a durable challenge to Anthropic’s production position.

The broader pricing trend also needs careful interpretation. Vercel’s official July index reported a 13.6% month-over-month decline in average price paid per token, alongside continued growth in token consumption and spending. That decline reflects a changing mix of models and workloads; it should not be treated as proof that every frontier model has become permanently cheaper.

The bottom line

GPT-5.6 Sol’s gateway discount lowers the cost of trying OpenAI and makes provider switching easier. That can help OpenAI gain evaluation traffic, developer familiarity, and potentially production share.

But the evidence does not yet show that OpenAI has displaced Anthropic where revenue matters most. Anthropic’s lead in gateway spending remains substantial, while Chinese open-weight models dominate much of the raw token market. The competitive outcome will depend less on the headline 50% discount than on whether Sol can deliver sustained quality and reliability at an attractive price after the promotion ends.