Choosing an AI model in 2024 is mostly a pricing exercise. The capabilities of the big three providers have converged, so the real question is what a million tokens costs you, and how fast that bill grows once you move from prototype to production. This guide compares the current price lists of Anthropic, OpenAI and Google, as of spring 2024.
What the flagship models cost
Anthropic’s top model, Claude 3 Opus, is priced at 15 dollars per million input tokens and 75 dollars per million output tokens. OpenAI’s GPT-4 Turbo sits at 10 dollars input and 30 dollars output. Google’s Gemini 1.5 Pro undercuts both at 7 dollars input and 21 dollars output for prompts over 128k tokens.
For most production workloads the mid-tier models are the sweet spot. Claude 3 Sonnet costs 3 dollars per million input tokens, GPT-4o is expected to land around 5 dollars, and Gemini 1.5 Flash is announced at under 1 dollar. These numbers move fast: every provider has cut prices at least once in the past twelve months.
The hidden costs nobody budgets for
Token prices are only part of the story. Rate limits force many teams onto higher pricing tiers before raw volume does. Output tokens are typically three to five times more expensive than input tokens, which penalizes long-form generation. And retries on timeouts silently multiply costs: a pipeline that retries twice on failure can spend forty percent more than its theoretical budget.
Context windows matter too. Claude 3 offers 200k tokens, GPT-4 Turbo 128k, and Gemini 1.5 Pro up to 1 million tokens in preview. A bigger window is not automatically cheaper: stuffing your whole knowledge base into every prompt costs more than a decent retrieval setup.
How to keep the bill predictable
Three habits keep AI spend under control. First, estimate before you launch: multiply expected input and output tokens by the current price list, and treat any job above your threshold as needing manual approval. Second, log real costs per task, not per month, so you can spot the one workflow that eats the budget. Third, re-evaluate your model choice quarterly. The price of intelligence has dropped roughly tenfold in eighteen months, and the model that was too expensive last year is probably the bargain of today.
Bottom line: in spring 2024, Gemini 1.5 Flash is the cheapest capable model, GPT-4 Turbo the middle ground, and Claude 3 Opus the premium option for hard reasoning tasks. Check the providers’ pricing pages before committing, because these numbers will not survive the year.
Leave a Reply