AI Model Prices Drop Sharply as Competition Increases

The cost of using advanced AI models has decreased substantially between 2020 and 2026, with several providers now offering lower pricing and a wider range of options for developers and businesses.

Meta launched its Muse Spark 1.1 model through the Meta Model API on July 9, 2026, setting input token prices at $1.25 per million and output tokens at $4.25 per million. This pricing is positioned below several other leading models, reflecting a broader trend of declining costs in the sector.

Anthropic’s Claude Opus 4.8 is priced at $5 per million input tokens and $25 per million output tokens, while SpaceXAI’s Grok 4.5 is available at $2 per million input tokens and $6 per million output tokens. These figures highlight a competitive environment where providers are adjusting their rates to attract users.

What the numbers show

• Muse Spark 1.1 output tokens cost $4.25 per million
• Claude Opus 4.8 output tokens are priced at $25 per million
• Frontier model token prices have dropped around 600-fold since 2020
• The Herfindahl-Hirschman Index for AI providers fell from 4,558 to 2,086 in three years

Media analysis indicates that Muse Spark 1.1’s output token price is about one-quarter of the output price for Claude Opus 4.8 and GPT-5.5. Codersera reporting also states that Grok 4.5’s output token price is roughly four times lower than that of Claude Opus 4.8, further illustrating the downward trend in costs.

Performance comparisons show that Muse Spark 1.1 scores 51 on the Artificial Analysis Intelligence Index, placing it three points behind Grok 4.5 and five behind Claude Opus 4.8. However, Muse Spark 1.1 leads on certain professional tool-use benchmarks, such as JobBench and MCP Atlas, according to media analysis.

Data from the Tiered Super-Moore’s Law report documents a dramatic reduction in token prices for frontier AI models, with an estimated 600-fold decrease over six years. This shift has coincided with a notable decline in market concentration, as measured by the Herfindahl-Hirschman Index, which dropped from 4,558 to 2,086 in three years.

These developments suggest that users now have access to a broader selection of AI models at lower costs. Providers are differentiating their offerings through pricing as well as performance on various benchmarks.

As the market for advanced AI models continues to evolve, pricing structures and model capabilities remain key factors influencing adoption and competition among providers.

* This article is based on publicly available information at the time of writing.