The report, released on June 26, emphasizes the substantial differences in compute requirements for various AI tasks. Notably, app and website creation, code debugging, spreadsheet analysis, and document generation are identified as the most demanding workloads.
These findings highlight a widening gap in the economics surrounding generative AI, where more complex tasks demand significantly greater computing resources — and correspondingly higher API costs — in contrast to simpler requests.
Also read: China, India see large companies lose market cap share, hinting at lag in AI raceWhat are AI tokens?
AI models do not comprehend or produce text in words or sentences. Instead, they interpret text as tokens—small segments that can comprise whole words, parts of words, punctuation, or numbers. For instance, the phrase “How are you?” would be segmented into various tokens prior to being processed by the model.
Every user prompt, along with every AI-generated response, utilizes tokens, making them the fundamental unit for assessing compute usage and API costs.
Why do some AI tasks cost more than others?
Anthropic calculates AI compute based on tokens, and since developers are charged per token, tasks that require more tokens inevitably incur higher costs.
The study revealed that app and website creation typically consumes around three to four times the tokens of an average Claude conversation. Other computationally intensive tasks include code debugging, data analysis, and document generation. In contrast, generating a simple explanation only requires about one-fifth of the median token count.
| AI task | Token use vs median conversation (Approx) | Relative API cost |
| App/website creation | 3–4x | Highest |
| Code debugging | 2–3x | Very high |
| Data/spreadsheet analysis | 2x | High |
| Document/report | 2x | High |
| Scripts/snippets | 1.5–2x | Above average |
| Plans/strategy | 1.5x | Above average |
| Analysis/summary | Around median | Moderate |
| Marketing content | Below median | Lower |
| Email drafting | 0.5x | Low |
| Explanation/answer | 0.2x | Lowest |
The relative API cost is based on Anthropic’s token distribution and API pricing. Actual costs may vary depending on the Claude model utilized.How much does Anthropic charge?
Anthropic charges API developers based on the number of tokens processed, pricing Claude Sonnet at $3 per million input tokens and $15 per million output tokens, while Claude Opus comes to $5 and $25 per million input and output tokens, respectively. Consequently, tasks that require more tokens can be several times more expensive than simpler AI operations.
The report also highlighted that tasks with higher value often need more compute. Conversations tied to higher-paying professions consumed approximately twice as many tokens as those linked to lower-paying roles, with around 44% of this variance attributed to the type of output generated by users.
In summary, requesting an AI to clarify a concept is relatively affordable, whereas asking it to create an application, debug code, or analyze large data sets is significantly more compute-heavy and costly, showcasing the stark differences in AI economics based on the task at hand.
Read more: Why the ₹15,000 smartphone is disappearing
