Curious about today's AI digest?ai-tldr.dev

Daily Digest

Gemini 4 Neon Cuts Prices as AI Margins Tighten

TechnologyMAJOR16m ago6 min read
Share
Gemini 4 Neon Cuts Prices as AI Margins Tighten

Google's Gemini 4 Neon, priced at $3 per million input tokens, lands as OpenAI and Microsoft also lower costs. The AI price war is squeezing model margins.

  • Google launched Gemini 4 Neon on Nov. 18 at $3 input and $15 output per million tokens, 25% below Carbon's scheduled list price.
  • OpenAI's GPT-6.2 Astra cut token prices 20%, while Microsoft says in-house models cut costs up to 89% versus OpenAI.
  • Alphabet (GOOGL) closed at $282.44, up 0.8%, on Nov. 18, and Microsoft (MSFT) rose 0.4% to $508.29.

Lead

Google released Gemini 4 Neon on Nov. 18, pricing the model at $3 per million input tokens and $15 per million output tokens. It sits between Hydrogen and Carbon in Google's lineup and is positioned as Carbon-level quality at Flash-level speed. The launch came a day after OpenAI introduced GPT-6.2 Astra and Sol with lower API pricing, and months after Microsoft began moving products onto its own models. Together the moves have turned "frontier-ish for less" into the dominant commercial theme in AI, and they put pressure on the margins of companies that sell model access.

What Is Gemini 4 Neon and How Is It Priced?

Gemini 4 Neon is a paid-only model for coding, analytical reasoning and enterprise workflows. It carries a 1 million-token context window and four thinking levels, from Low to Extra High. Thinking tokens are included in the output price, and context caching costs $0.15 per million tokens.

Its pricing is cheaper only on a forward-looking basis. Carbon costs $2 input and $10 output through Nov. 30 under a promotional rate, then rises to $4 and $20. Neon undercuts that later list price by 25% but costs more than Carbon does today. It is twice the price of Hydrogen. It is live for AI Pro and Ultra subscribers in the Gemini app and on Google's Antigravity paid tiers, and it replaces Argon.

On benchmarks, the model scores 61 on the Artificial Analysis Intelligence Index, third overall. It ranks first on the Legal Research Bench, Tax Agent Bench and Zapier's AutomationBench. Independent testing shows it one point ahead of Carbon with lower measured cost and faster completion.

How Did OpenAI and Microsoft Respond on Price?

OpenAI lowered prices on its top model, while Microsoft is cutting its own costs by relying less on outside suppliers. GPT-6.2 Astra is priced at $8 per million input tokens and $40 per million output tokens, a 20% reduction from GPT-6 Astra. Cached reads fell 60%, from $1.00 to $0.40. GPT-6.2 Sol holds at $2 input and $10 output, unchanged from its predecessor and below Neon's $3 input rate. Long-context requests cost more on both models: $16 input and $60 output for Astra, and $4 and $15 for Sol.

Microsoft's route is different. At Build 2026 it announced seven in-house models, and it has replaced OpenAI and Anthropic models in parts of Copilot, including Excel and Outlook. The company says its production metrics show GPU costs falling by up to 89% against OpenAI, and its MAI models run about 85% cheaper in PowerPoint. Because Microsoft avoids licensing fees and revenue-share payments, it keeps more of the cloud margin while passing savings to enterprise buyers.

Why Is the Price War Squeezing AI Model Margins?

Falling per-token prices squeeze margins because the cost of serving each token drops more slowly than the price buyers will pay. Model builders face heavy fixed costs in training and chips, so each price cut reduces revenue without an equal reduction in serving cost.

Three dynamics are visible in the current round:

  • Tiered lineups. Each vendor now sells a premium model and a cheaper model near the frontier, and customers can move down the ladder with little loss of quality. GPT-6.2 Sol has been measured as the cheapest route past 55 on the Artificial Analysis index, at about $1.01 per index task at maximum effort.
  • Caching discounts. Cut rates on cached input, such as OpenAI's 60% reduction, favor high-volume enterprise users and lower effective revenue per request.
  • Captive demand. Microsoft's shift to its own models removes a large customer from third-party suppliers, which pressures OpenAI's volumes.

Investors have not repriced the large platforms on the news. Alphabet shares rose 0.8% on Nov. 18 and Microsoft added 0.4%, against a 0.42% gain for the Nasdaq. That day also brought Anthropic's IPO pricing, in which Google committed about $10 billion, a development that drew more attention than the model pricing itself.

What Does This Mean for AI Stocks?

For investors weighing ai stocks, the price war changes where returns are likely to accrue. Cheaper inference favors companies that own their own silicon, cloud capacity and distribution, and it weighs on standalone model vendors that rent compute. Alphabet, with its TPU lineup and Gemini subscriptions, and Microsoft, with in-house models inside Office and Azure, can absorb lower per-token prices because they capture value elsewhere in the stack. Pure-play model providers lack that offset.

Lower prices also expand demand. Enterprises that held back agent and coding deployments over cost now face a lower entry point, which raises token volumes even as revenue per token falls.

Outlook

The next test is Dec. 1, when Carbon's promotional pricing ends and Neon's 25% discount to the new list price takes effect. Pricing decisions from OpenAI on Sol and from Microsoft on its MAI line will show whether the cuts stay targeted or become broader. Volume growth will determine whether lower prices shrink margins for the largest platforms or only for smaller model vendors.

Mentioned tickers: GOOGL, GOOG, MSFT

The Daily Briefing

Every story that moved the market, every weekday.

Market news - the major stories only, free, and one email a day.

One email a day. Unsubscribe anytime.