Claude API pricing
Official model token pricing table.
← Back to Compute College
Compute College
Why output tokens cost more than input tokens
Learn why output tokens usually cost more than input tokens and how generation cost affects model serving economics, AI agents, and inference spend.
Output token pricing definition
Output tokens often carry a higher listed API price than input tokens because generation is performed sequentially during the response, tying up serving resources as each next token is produced; actual pricing is set by each provider.
Memory trick: Input loads the workbench; output keeps the generator working token by token.
Why it matters
Why output tokens drive serving cost
Output-heavy coding, reasoning, and agent workflows can create much larger serving bills than short responses, so token mix matters for model adoption and AI compute demand.
- Output-heavy coding, reasoning, and agent work can dwarf short-response bills.
- Generation runs sequentially, tying up serving resources per token produced.
- So token mix, not request count, often drives serving demand.
Simple example
Anthropic lists Claude Opus 4.8 regular usage at $5 per million input tokens and $25 per million output tokens. At those listed rates, an illustrative request with 100,000 input tokens and 20,000 output tokens costs $0.50 for input and $0.50 for output: much fewer output tokens contribute the same spend.
- Providers commonly list output tokens at several times the input rate.
- At $5/M input and $25/M output, far fewer output tokens can match input spend.
- A short prompt with a long answer can be an output-dominated bill.
Example figures are illustrative calculations, not current quoted market prices.
Current example
Current pricing example: Claude Opus 4.8
Anthropic’s official pricing documentation lists Claude Opus 4.8 at $5 per million base input tokens and $25 per million output tokens. Anthropic’s Opus 4.8 release also lists fast mode at $10 per million input tokens and $50 per million output tokens. Last checked: June 1, 2026.
Pricing is current-source information and should be checked again before making a procurement decision.
Market signal
How to read the market signal
If buyers adopt agents or long-form reasoning workflows that produce more output, model-serving demand and spend can grow even when request count or input volume appears stable.
- Adoption of agents or long-form reasoning grows output volume.
- Serving demand and spend can rise even when request count looks stable.
- Watch output share, not just total requests, when reading demand.
Market read: a shift toward output-heavy workloads (agents, long-form reasoning) can grow serving spend even with flat request counts. Figures here are illustrative unless explicitly sourced and dated — see our methodology.
Common mistake
Do not estimate a workload from input tokens alone, and do not assume every provider or model has the same input-to-output pricing relationship.
Practical takeaway
What you can do with this
Track input and output token volumes separately, use current official pricing, and model how answer length, reasoning, and agent steps change cost per completed task.
- Track input and output token volumes separately.
- Use current official pricing, since input-to-output ratios vary by provider.
- Model how answer length, reasoning, and agent steps change cost per task.
Decision check: are you tracking output tokens separately and using current per-provider rates, rather than estimating from input volume alone?
Compute College
Follow model releases as market signals
Follow model releases as AI compute market signals in the ComputeTape Morning Brief.
Compute College track
Model Benchmarks & AI Compute Economics
Step 22 of 23: Why output tokens cost more than input tokens