GPT-6.2, Gemini 4 Neon and Qwen4-Max: The AI Model Price War Cutting Agent Costs in a Week

J
By jenifer — 22/11/2026No Comments5 Mins Read
GPT-6.2, Gemini 4 Neon and Qwen4-Max: The AI Model Price War Cutting Agent Costs in a Week

It took less than a week for the price of frontier AI to fall off a cliff. Between 17 and 18 November, OpenAI launched GPT-6.2, Google launched Gemini 4 Neon, and Alibaba's Qwen4-Max arrived with a price tag that undercut them both. Add Microsoft's seven new MAI models and its Maia 300 chip, and the message is unmistakable: the AI model price war has begun. This is not a story about which model is smartest. It is a story about cost, and cost is what decides which AI tools businesses can actually afford to use at scale.

What Happened in the AI Model Price War This Week?

The launches came so fast that they are best understood together, as a single competitive shock — four major launches in a matter of days, each one cheaper than the last. That is not a coincidence; it is a price war.

  • 17 November — GPT-6.2: OpenAI launched two versions, Astra and Sol. Astra, the flagship, is priced at $8 per million input tokens and $40 per million output tokens. Sol, the lighter model, costs just $2 per million input tokens and $10 per million output tokens. We compared Astra with Google's flagship in our GPT-6.2 Astra vs Gemini 4 comparison.

  • 18 November — Gemini 4 Neon: Google launched its new mid-tier model at $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window — a huge amount of working memory for a model at this price. Full details are in our Gemini 4 Neon explainer.

  • Qwen4-Max: Alibaba's new model arrived with a 2.4 trillion-parameter design priced at $1.5 per million input tokens and $4.5 per million output tokens — the cheapest of the new flagship-class models by a wide margin.

  • Microsoft's answer: Microsoft launched seven MAI models of its own, alongside the Maia 300 chip, signalling that it intends to compete on both software and hardware.

How Do GPT-6.2, Gemini 4 Neon and Qwen4-Max Compare on Price?

The headline numbers tell the story. Prices are quoted per million tokens — a token is roughly a word or part of a word, and output tokens cost more because generating text takes more computing power.

  • GPT-6.2 Astra: $8 input / $40 output per 1m tokens

  • GPT-6.2 Sol: $2 input / $10 output per 1m tokens

  • Gemini 4 Neon: $3 input / $15 output per 1m tokens

  • Qwen4-Max: $1.5 input / $4.5 output per 1m tokens

The gap between the most and least expensive is striking. Qwen4-Max's output price is nearly nine times cheaper than GPT-6.2 Astra's. And Sol — OpenAI's own cheaper model — costs a quarter of its flagship sibling. Price is not everything. The cheapest model is not automatically the best choice for every job, and the differences in capability between these models are real. But the direction of travel is clear: the cost of a given level of AI capability is falling fast, and falling across every provider at once.

Why Are AI Model Prices Falling So Fast?

Three forces are driving prices down at the same time.

  1. Fierce competition: OpenAI, Google, Alibaba and Microsoft are all fighting for the same customers. When one provider cuts prices, the others have to follow — or lose business.

  2. Cheaper, more efficient models: The new models are designed to do more with less computing power. Techniques that make models smaller and faster without losing much capability are now standard practice, and the savings are being passed on to customers.

  3. The race for developers: The company that wins over the developers building AI agents wins the next decade. Low prices are the fastest way to get developers to try a model, build on it, and stay.

There is also a bigger shift underneath it all. AI models are becoming a commodity — a standard product where providers compete mainly on price and reliability, not just on raw intelligence.

What Do Cheaper Models Mean for Businesses Building AI Agents?

AI agents — software that can carry out multi-step tasks on its own, from answering customer emails to analysing documents — have been held back by cost. An agent that makes hundreds of model calls a day can run up a large bill very quickly. That changes now.

  • More experimentation: When each test costs a fraction of what it did, teams can try more ideas, build more prototypes and find what actually works.

  • Smaller businesses can compete: The price war lowers the barrier to entry. A startup with a small budget can now use models that were, until recently, the preserve of companies with deep pockets.

  • The agent gold rush accelerates: Expect a wave of new AI agent products in the coming months, as the cost of building them falls and the number of businesses that can afford them grows.

What Should Businesses Watch Before Switching Models?

Cheaper is not always better, and switching models is not free. Before moving an AI agent to a new, cheaper model, test it on your own tasks — benchmark scores do not tell you how a model performs on your specific work. Watch the total cost, not just the per-token price: the number of calls an agent makes matters just as much. And check reliability and support, because a cheap model that goes down during business hours is more expensive than a pricier one that stays up.

Final Thoughts

A week ago, running a serious AI agent was an expensive proposition. Today, the same work costs a fraction of what it did — and the price war is only just beginning. GPT-6.2, Gemini 4 Neon and Qwen4-Max have turned the AI model market into a buyer's market, and the winners will be the businesses that move fastest to take advantage of it.

Disclaimer: Prices quoted are as announced at launch and may change. This article is for information only and is not purchasing advice. Always check current pricing on the provider's official website before making a decision.

CategoryDetails
TopicTechnology
Authorjenifer
Published22/11/2026
Read TimeNot set
J

jenifer

Read more articles by this author and explore related coverage across the site.

View All Posts