Verified 2026-10-08 · sourced from Anthropic
100K Claude Haiku 5.5 Tokens — Cost Breakdown
Use this guide to benchmark budgets for 100,000 tokens. Standard pricing is $0.10 per million input tokens and $0.50 per million output tokens. Cached input, when available, reduces costs significantly for repeated contexts.
How much do 100K tokens cost?
- 100K input only
- $0.0100
- 100K output only
- $0.0500
- 50,000 input + 50,000 output
- $0.0300
USD, standard uncached token charges. Output-only is a billing comparison across supported requests, not a single-response limit. Cache writes, tools, taxes, and other fees are excluded; prompt length and service mode can change rates.
Reference rates: Standard · ≤100K input · global
USD per million text tokens. Verified 2026-10-08 · Official source
| Mode / prompt size | Input | Cache read | Cache write | Output |
|---|---|---|---|---|
| Standard ≤100,000 input | $0.1 | $0.01 | $0.125 (5m) / $0.2 (1h) | $0.5 |
| Standard >100,000 input | $0.5 | $0.05 | $0.625 (5m) / $1 (1h) | $2.5 |
| Batch ≤100,000 input | $0.05 | $0.005 | $0.0625 (5m) / $0.1 (1h) | $0.25 |
| Batch >100,000 input | $0.25 | $0.025 | $0.3125 (5m) / $0.5 (1h) | $1.25 |
1,000,000 context tokens · max output 128,000. Local token counts are estimates; use the provider’s usage counts for an invoice estimate.
Text-only estimate. Local token counts are approximate; use provider-reported usage for billing. Tool, image, audio and video charges are excluded.
Up to 100,000 input tokens uses the lower tier. Above 100,000, the full request uses higher input, cache and output rates, not just the excess tokens. The threshold is per request; daily call volume does not change it.
For prompt-size tier selection, total input includes uncached, cache-read and cache-creation tokens, using the provider’s total-input definition. This calculator applies one selected input billing operation to all input; it does not split mixed cache usage.
US-only inference adds 10% to token rates and is excluded from this global-endpoint estimate. Newer Claude tokenization differs from the local estimator; no fixed conversion factor is applied.
Billed output includes reasoning/thinking tokens that cannot be inferred from pasted final text; use API-reported output usage for accurate billing.
Batch: Batch halves token rates and stacks with cache read/write pricing. Cache hits in asynchronous batches are best effort, not guaranteed. Cache writes use separate 5-minute and 1-hour retention rates.
Scenario breakdown
Cost estimates for different input/output distributions using 100K total tokens.
| Scenario | Tokens in | Tokens out | Standard cost | Cached cost |
|---|---|---|---|---|
Balanced conversation 50% input · 50% output | 50,000 | 50,000 | $0.0300 | $0.0255 |
Input-heavy workflow 80% input · 20% output | 80,000 | 20,000 | $0.0180 | $0.0108 |
Generation heavy 30% input · 70% output | 30,000 | 70,000 | $0.0380 | $0.0353 |
Cached system prompt 90% cached input · 10% fresh output | 90,000 | 10,000 | $0.0140 | $0.0059 |
Workload multipliers
Each run uses 100K tokens split 50/50 between input and output. Daily cost multiplies that request cost; monthly assumes 30 days. Cached columns keep the same split and assume all input qualifies for cache reads, excluding cache writes.
| Profile | Runs/day | Tokens/day | Daily cost | Monthly cost | Cached daily | Cached monthly |
|---|---|---|---|---|---|---|
| Single workload | 1 | 100,000 | $0.0300 | $0.900 | $0.0255 | $0.765 |
| Daily workload (10 runs) | 10 | 1,000,000 | $0.300 | $9.00 | $0.255 | $7.65 |
| Team workload (100 runs) | 100 | 10,000,000 | $3.00 | $90.00 | $2.55 | $76.50 |
Frequently asked questions
What is the standard cost of 100K Claude Haiku 5.5 tokens?
100K input-only tokens cost $0.0100; output-only tokens cost $0.0500; 50,000 input plus 50,000 output tokens cost $0.0300. These are USD token charges at standard rates, excluding cache writes, tools, taxes, and other fees. Output-only is an accounting example across supported requests, not a promise of one response that long.
What happens if cached input is enabled?
With cached contexts, the same 100K tokens drop to $0.0255 when all input is an eligible cache read at $0.010 per million tokens.
How many requests does 100K tokens cover?
If your prompts average around 3,000 tokens per call, 100K total tokens cover about 33 requests.
How fresh is the pricing information?
Prices are taken from https://platform.claude.com/docs/en/about-claude/pricing and were last verified on 2026-10-08. This guide uses the recorded catalog rates; check the official source before budgeting.