The AI Price War Has Started

The beginning of autumn brought an unexpected gift for AI users: stronger models at lower prices.

Anthropic launched Opus 5.5 with improved performance at 80% of the cost of Opus 5. OpenAI responded with two new models, GPT-6 Sol and GPT-6 Luna, while also cutting prices by 50% across its model lineup.

These releases are more than routine product updates. They signal that the AI market is entering a new phase, where model quality and pricing are becoming increasingly competitive. For organizations building AI workflows, the question is no longer simply which model is most capable. It is also which model delivers the right level of reasoning for the lowest practical cost.

Anthropic Makes a Strong Move With Opus 5.5

Anthropic’s Opus 5.5 is positioned as a direct upgrade for existing Opus users.

The new model costs $4 per million input tokens and $20 per million output tokens. That is down from $5 per million input tokens and $25 per million output tokens for Opus 5. The reduction is meaningful for teams running large-scale workflows, especially those that process substantial amounts of text or generate long responses.

The pricing would already be notable on its own, but Anthropic also claims a performance improvement. According to the company’s published benchmarks, Opus 5.5 outperforms Fable 5.1 across every evaluation included in the comparison.

That result is particularly striking because Fable 5.1 was released only three weeks earlier. A model that was recently presented as the newest high-performance option has already been surpassed by another model from the same company.

For organizations already using Anthropic, the decision appears relatively straightforward. Opus 5.5 offers a newer model, stronger reported benchmark performance, and a lower price than its predecessor.

OpenAI Targets Everyday Workflows

OpenAI’s response focuses on a different part of the market.

The company launched GPT-6 Sol and GPT-6 Luna. These models are not positioned as frontier-level systems. That role belongs to Astra. Instead, Sol and Luna are designed to serve as efficient daily drivers for routine work.

• GPT-6 Sol is priced at $2 per million input tokens and $10 per million output tokens.
• GPT-6 Luna costs just $0.10 per million input tokens and $0.50 per million output tokens.

Those prices create a much wider range of options for teams building AI-powered products. Organizations do not need to use a frontier model for every task. Many workflows require reliable summarization, classification, extraction, drafting, or basic reasoning rather than the highest level of available intelligence.

For those use cases, a lower-cost model can produce better overall economics.

The goal is not always to maximize intelligence. It is often to maximize useful output per dollar.

The Right Model Depends on the Workflow

The latest releases make one point clear: model selection is becoming more nuanced.

Some workflows require advanced reasoning, complex planning, or the ability to handle difficult and ambiguous instructions. In those cases, the additional cost of a more capable model may be justified.

Other workflows are repetitive and predictable. They may involve processing large volumes of documents, organizing information, responding to common requests, or completing structured tasks. For these applications, a smaller and more efficient model may be the better choice.

This means that companies should evaluate models based on the actual requirements of each workflow rather than choosing one model for everything.

A mature AI strategy may use several models at once. A high-capability model can handle the most demanding tasks, while more affordable models manage routine workloads. The result can be a better balance between quality, speed, and operating cost.

A New Era of Competition

The timing of these releases is also revealing. Anthropic and OpenAI launched their latest models on the same day, yet neither company could include the other’s newest model in its published benchmarks.

That limitation is amusing, but it also highlights how quickly the market is moving. Model releases are arriving so rapidly that even major companies are competing against systems that were not available when their evaluations were designed.

The competitive pressure is likely to continue. As model capabilities improve, providers will need to justify their prices with measurable performance. At the same time, price reductions will make advanced AI accessible to more teams and allow existing customers to scale their usage.

The result is a market where progress is being measured in two dimensions at once: how intelligent a model is and how efficiently it delivers that intelligence.

The AI price war has started. For businesses building with these systems, that is good news.

More capable models, lower prices, and a broader range of options give teams greater flexibility in designing AI workflows that fit their needs