On 6 October 2026, the cost of generating one million tokens of output varied significantly among eighteen AI models from companies like OpenAI, Anthropic, Google, Mistral AI, DeepSeek, and xAI. According to data collected by ActuIA, the prices ranged from 0.50 dollars for OpenAI's GPT-6 Luna to 50 dollars for GPT-6 Astra and Anthropic's Claude Fable 5.1. The figures were converted into euros using the European Central Bank’s exchange rate of 1 euro for 1.1204 dollars, which was the most recent rate at the time of the recording. The data was gathered between 0 h and 1 h UTC on that day. The prices listed apply to the standard rate, which assumes a "short context" of up to 272,000 input tokens for OpenAI models and 200,000 tokens for Google's Gemini 3.1 Pro and Grok 4.7. Some models, such as xAI’s Grok 4.7, charge a higher rate for all tokens in a request if the prompt exceeds the threshold. Google applies a higher rate for prompts exceeding 200,000 tokens, while OpenAI does not clearly explain how the switch to the higher threshold affects billing. The "input from cache" column indicates the cost of reading tokens from a request cache, without including the cost of writing new data. The "lots" column refers to deferred processing via Batch API, which typically offers a discount in exchange for a longer response time. The full list of models and their respective rates, in dollars per million tokens, included several variants from OpenAI, Anthropic, Google, Mistral AI, and DeepSeek, with some models offering different prices during peak and off-peak hours. When converted to euros, the prices showed similar patterns, with GPT-6 Astra costing around 8.93 euros for input, while GPT-6 Luna was as low as 0.089 euros for input. The discount for tokens read from the cache ranged from 75% for Grok 4.7 to 98% for DeepSeek's off-peak hours. However, these discounts apply only to identical requests where the initial instructions and reference documents are the same across calls. Several new models were launched in the months leading up to the recording, including Anthropic’s Claude Fable 5.1 and DeepSeek’s V4.1 Flash, which introduced lower pricing. OpenAI and Mistral AI also released new models, while Google introduced Gemini 3.8 Flash at a reduced rate. Additional charges, such as storage rental and data residency, were noted, with regional access points adding 10% to the cost. Some models, like DeepSeek, offer lower rates during off-peak hours, while others, like Anthropic, allow for cumulative discounts on batch processing, caching, and data residency.