As of mid-2026, Claude API rates per million tokens are: Claude Haiku 4.5 at $1.00 input / $5.00 output, Claude Sonnet 4.6 at $3.00 / $15.00, and Claude Opus 4.8 at $5.00 / $25.00. Output tokens cost 5× the input rate across all current models. Batch processing halves all rates, and prompt caching reduces cached input by up to 90%.

A token is roughly 3–4 characters, or about ¾ of a word. One million tokens is approximately 750,000 words — around 1,500 pages of text. Anthropic bills separately for input tokens (your prompts, system instructions, documents, and context) and output tokens (Claude’s responses). A typical API call might use 1,000–5,000 input tokens and 200–1,000 output tokens.

The Anthropic Batch API provides a flat 50% discount on all input and output tokens. Jobs are submitted asynchronously and completed within 24 hours. This is ideal for document processing pipelines, content generation, data classification, nightly analytics, and any workload where an immediate response is not required. Batch and prompt caching discounts can be combined for even greater savings.

Prompt caching reduces cached input token costs by up to 90%. It works by storing frequently reused content — such as system prompts, reference documents, and tool definitions — so subsequent calls can read from cache instead of reprocessing. Cache writes cost 1.25× input price (5-minute TTL) or 2× (1-hour TTL), but cache reads are 90% cheaper than fresh input. A production app with a 3,000-token system prompt sent across 10,000 daily requests can save thousands of dollars per month.

Claude Sonnet 4.6 ($3/$15 per MTok) is the recommended default for most production workloads — coding, analysis, writing, RAG pipelines, and customer-facing applications. Claude Haiku 4.5 ($1/$5) is ideal for high-volume, latency-sensitive tasks like classification and routing. Claude Opus 4.8 ($5/$25) is best reserved for the most complex reasoning, agentic engineering, and tasks that genuinely require maximum capability. A common pattern is 70% Haiku / 20% Sonnet / 10% Opus to balance cost and quality.

The Claude API is pay-as-you-go with no monthly subscription fee. You pay only for the tokens you consume. New accounts start with prepaid credits. This is entirely separate from Claude.ai consumer plans (such as Claude Pro at $20/month or Claude Max), which provide chat access and do not include API usage. Higher-volume accounts can switch to monthly invoicing.

Claude Opus 4.8 Fast Mode provides significantly faster inference at $10 input / $50 output per million tokens — 3× cheaper than the previous Opus 4.7 Fast Mode ($30/$150). It is currently available in research preview and is designed for agent workflows and coding tasks where response speed is critical. Standard Opus 4.8 at $5/$25 remains available for non-time-sensitive requests.

Yes — emphatically. Opus 4.8 costs $5/$25 per million tokens versus $15/$75 for the legacy Opus 4.1, a 67% price reduction, while delivering meaningfully better performance on reasoning and coding benchmarks. Any remaining Opus 4.1 workloads should be migrated as a priority. Note that migrating from Opus 4.6 to 4.7 or 4.8 involves a tokenizer change that may increase token counts by up to 35% — benchmark your specific workloads before migrating to account for this.