Cost to Process 100,000 Tokens

100,000 tokens is a practical benchmark for understanding single-request costs. In English text, 100K tokens represents approximately 75,000 words — a full-length novel, a 250-page technical manual, or a moderately-sized codebase. For shorter requests — typical chatbot messages average 500–2,000 tokens — 100K tokens represents 50–200 individual conversations.

Per-request cost for common workloads. A typical customer service message with a 1,000-token system prompt, 500-token user message, and 300-token response totals 1,800 tokens. At GPT-4o Mini ($0.15/M input, $0.60/M output), that single interaction costs $0.000330 — less than a tenth of a cent. At GPT-4o ($2.50/M input, $10/M output), the same interaction costs $0.005500. The per-request difference is small, but at 100,000 requests per day it becomes $330/day versus $550/day — a $220 daily difference from a single model choice.

Long-context use cases. Models with large context windows — Gemini 2.5 Pro (1M tokens), Claude Opus (200K), GPT-4o (128K) — enable processing entire documents in a single API call. Analyzing a 100,000-token legal contract costs $0.125 in input tokens on Gemini Flash, $0.25 on Gemini 2.5 Pro (under 128K threshold), and $0.30 on Claude Sonnet. For single-document analysis, the absolute cost is modest even on premium models. The economics only become critical at scale — processing thousands of documents per day.

Choosing between models at similar price points. Several models cluster around similar price points for 100K tokens. When models are similarly priced, the deciding factors shift to: latency (streaming speed), context window size, structured output support, function calling reliability, and your team's familiarity with the API. Price per token is the starting filter, not the final decision criterion. Use the comparison tool to evaluate models side-by-side across the dimensions that matter for your specific workload.