Animated token-price trend visualization comparing OpenAI and Anthropic models

THE AI PRICE CRASH: HOW TOKEN COSTS PLUMMETED 73% AND OPENAI & ANTHROPIC SWAPPED ROLES OVER 4 YEARS A visualization of token price trends from November 2022 to September 2026 reveals the real economics of neural networks and the fierce war between Claude and GPT. Key figures from the four-year race: Anthropic’s flagship models dropped in price by 73%: The blended price per 1 million tokens fell from $30 (Claude 3 Opus) to $8 (Opus 5.5). GPT-5 became 11 times cheaper than GPT-4: The price plummeted from $37.50 per 1 million tokens for GPT-4 to $3.44 for GPT-5 (though the top-tier GPT-6 Astra shot back up to $20). Cheapest input on the market ($0.05): GPT-5-nano hit a new price floor ($0.05 per 1 million input tokens, compared to $1 for Haiku 4.5). Extreme caching (discounts down to 2.5%): Reading from the cache with Fable 5.1 costs just $0.25, versus $10 for standard input. Release volume: Anthropic overtook OpenAI in the number of models launched-35 versus 27. How are developers saving on AI in 2026? 1. Batch API: Offers discounts of up to 90% from both vendors for asynchronous requests. 2. Context caching: The cost of re-reading data dropped from 96% of the base price (GPT-6a) to a record low of 2.5% (Fable 5.1). 3. The tokenizer trap: Claude’s tokenizer counts 30% more tokens for the same text compared to OpenAI’s solutions. 4. The cost of "reasoning": Hidden "reasoning" tokens are billed at the high rate used for output tokens. Which strategy is winning-OpenAI’s ultra-cheap mass-market approach or Anthropic’s optimized context? 👇

— Yumslef (@Yumslef) Sep 24, 2026