Claude API Pricing Guide, August 2026

Current Claude API Pricing, August 2026

Anthropic sells the Claude 5 generation alongside the older 4.x models. The Claude 5 generation (Fable 5, Opus 5, Sonnet 5) leads the table, and the older Opus 4.x, Sonnet 4.x, and Haiku 4.5 models stay on sale at their existing rates. Pricing is based on input and output tokens separately, which means the cost of a request depends on both how much context you send and how long the response is. All prices below are per one million tokens.

These prices apply to the standard real-time API. Anthropic states that "Claude 4.6 and later models and Claude Mythos Preview include the full 1M token context window at standard pricing." Prompt caching reduces the cost of repeated prefixes in your prompts. Per the pricing page, a cache read costs "0.1x base input price", a 5-minute cache write costs "1.25x base input price", and a 1-hour cache write costs "2x base input price". Cached input tokens therefore cost 90% less than standard input tokens, which makes caching effective for system prompts and few-shot examples that stay constant across requests.

Sonnet 5 stays at $2 input and $10 output. Anthropic launched Sonnet 5 with introductory pricing that was due to rise on September 1, 2026. That increase was cancelled. The pricing page now reads: "The $2/$10 per million input/output token pricing for Claude Sonnet 5, announced at launch as introductory pricing through August 31, 2026, is now the standard price. The previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur."

Price your own workload

Enter your token counts. Costs are computed in your browser from the same published rates shown in the table above, and are computed from the same rates shown above.

Cached input is charged at 10% of the standard input rate, the 90% reduction stated above. Tick the batch box to halve every figure, which is the Batch API discount. The default figures are the 10,000-conversations-a-day example used further down this page, at 1,000 input and 1,000 output tokens per conversation.

When to Use Each Claude Tier

Choosing the Claude model is usually the largest cost lever. Running Fable 5 when Haiku would suffice can inflate your bill by 10x. Here is a practical breakdown of when each tier makes sense.

Claude Haiku 4.5 is ideal for high-volume, low-complexity tasks: content classification, entity extraction, simple question answering, data validation, and routing queries to more expensive models. At $1.00 per million input tokens, one dollar buys one million input tokens. Haiku responds quickly, which suits latency-sensitive applications.

Claude Sonnet 5 is the workhorse for production applications. It handles multi-step reasoning, code generation, summarization, and conversational AI at a fraction of the Opus and Fable cost. Most teams should default to Sonnet and only escalate when Sonnet measurably underperforms. At $2/$10 per million tokens, a chatbot handling 10,000 conversations per day at 1,000 input and 1,000 output tokens each runs about $3,600/month. That is 300,000 requests, 300M input tokens at $2 per million is $600, and 300M output tokens at $10 per million is $3,000. Output tokens dominate the bill on every Claude tier, which is why trimming response length saves more than trimming prompts. Put your own numbers into the calculator above.

Claude Opus 5 is reserved for tasks that need deeper reasoning: complex code architecture decisions, nuanced legal or medical text analysis, multi-document synthesis, and advanced mathematical proofs. At $5/$25 it costs the same as Opus 4.8 and Opus 4.7. Claude Fable 5, at $10/$50, is the most expensive current model and doubles the Opus rate on every token. Note that Anthropic says "Claude 4.7 and later models and Claude Mythos Preview use a newer tokenizer" and that "This tokenizer produces approximately 30% more tokens for the same text". So effective cost per request on those models can be higher than per-token pricing suggests.

Cost Optimization Tips

There are several strategies to reduce your Claude API bill without sacrificing quality. First, implement a model routing layer that sends simple requests to Haiku and only escalates to Sonnet, Opus, or Fable based on complexity signals.

Second, use Anthropic's Batch API for non-real-time workloads. The Batch API processes requests asynchronously and, per the pricing page, offers "a 50% discount on both input and output tokens". This makes Sonnet 5 batch processing cost $1/$5 per million tokens and Sonnet 4.6 batch processing $1.50/$7.50.

Third, use prompt caching for repetitive system prompts. Cached input tokens cost 90% less. If your system prompt is 4,000 tokens and you send 100,000 requests per day, caching saves substantially on Sonnet 5. Fourth, minimize output tokens by instructing the model to be concise and using structured output formats like JSON, which tend to be more token-efficient than natural language responses.

Finally, consider the total cost of ownership. While Haiku is the cheapest per token, a task that requires three Haiku retries to get right may cost more than a single Sonnet call. Track your success rates per model tier and optimize for cost-per-successful-completion rather than raw token cost.

Related Guides

Built by Michael Lip. Prices last verified 2026-08-29 against the Anthropic pricing page.