SILICON VALLEY — Prices for artificial intelligence models from leading U.S. laboratories have decreased by almost 25 percent since mid-July, according to Silicon Data’s token price index. The decline reflects pricing adjustments by OpenAI and Anthropic alongside increased market competition from Chinese developers.

OpenAI reduced the price of its GPT-5.6 Luna model by 80 percent. In May 2026, the company introduced the model at a rate of $1 per million input tokens and $6 per million output tokens. By July 2026, those rates fell to $0.20 per million input tokens and $1.20 per million output tokens.

Anthropic launched the Claude Opus 5 model in July 2026. The company priced the new system at $5 per million input tokens and $25 per million output tokens. This rate is half the price of Anthropic's Fable 5 model.

The broader pricing environment shows additional reductions across product lines. The price of OpenAI’s GPT-5.6 Terra was reduced from $2.50 to $2 per million input tokens and from $15 to $12 per million output tokens in July 2026. Anthropic had originally planned to increase the price of its Sonnet 5 model in September 2026 but canceled the increase.

U.S. corporations are integrating foreign alternatives to manage expenses. DoorDash and Airbnb have started using Chinese-made AI models to reduce costs. DoorDash stated in a social media post that it has begun using Kimi K2.6 from Beijing-based Moonshot AI for lower-level tasks.

What's New

Additional reporting details the specific timeline of these product launches and price adjustments. In June 2026, Anthropic released Fable 5, a high-performance model that was priced at $10 per million input tokens and $50 per million output tokens, setting the stage for the subsequent launch of Claude Opus 5 at half the cost. Claude Opus 5, the latest model from Anthropic, was launched in July 2026 and is priced at $5 per million input tokens and $25 per million output tokens, which is half the price of its predecessor, Fable 5.

Further context includes earlier coverage on 2026-07-24 regarding how a Chinese AI model stopped OpenAI’s ‘unprecedented’ cyber attack. Research titled Sticky Prices: An Empirical Assessment of Alternative Models provides additional academic framing for these market shifts.

How Sources Differ

Reports present differing figures regarding the performance and pricing of the GPT-5.6 Luna model. One source states that in June 2026, OpenAI announced that GPT-5.6 Luna would outperform Fable 5 at an estimated cost per task nearly 99% lower, as measured by Agents’ Last Exam, showing the performance-to-cost advantage of the new pricing model. Another source states that OpenAI reduced the price of its GPT-5.6 Luna model by 80 percent.

Descriptions of the Claude Opus 5 launch and pricing also vary in detail. One account notes that Claude Opus 5, the latest model from Anthropic, was launched in July 2026 and is priced at $5 per million input tokens and $25 per million output tokens, which is half the price of its predecessor, Fable 5. Another source simply states that Anthropic launched the Claude Opus 5 model. A third source specifies that Anthropic priced Claude Opus 5 at $5 per million input tokens and $25 per million output tokens.

Benchmark assessments of the Luna model offer contrasting cost comparisons. One report indicates that in June 2026, OpenAI announced that GPT-5.6 Luna would outperform Fable 5 at an estimated cost per task nearly 99% lower, as measured by Agents’ Last Exam, showing the performance-to-cost advantage of the new pricing model. Another finding from Artificial Analysis states that OpenAI’s GPT-5.6 Luna at max effort performed similarly to DeepSeek’s V4 Flash at max effort but cost just under twice as much per task.

Why It Matters

These pricing dynamics occur as both major U.S. labs prepare for public markets. OpenAI and Anthropic have filed confidentially for initial public offerings. Reports indicate OpenAI is targeting a valuation of up to $1 trillion for its IPO. Anthropic was valued at $965 billion in its latest private funding round.

The convergence of lower prices and high valuations suggests a market where cost efficiency drives adoption ahead of potential listings. The cancellation of planned price increases and the rapid deployment of cheaper models indicate that competitive pressure from international providers like Moonshot AI is shaping corporate strategy. This environment forces established players to balance revenue goals with the need to retain enterprise customers who are increasingly willing to switch providers for better unit economics.