2026 pricing reference

How much does the Claude API cost? 2026 reference.

Anthropic prices the Claude API per million tokens of input and output, with separate rates for each model tier. Add prompt caching and batch processing and the realized cost for a typical B2B workload is materially lower than the headline rate.

Short answer

API pricing is per million tokens of input and output, set per model tier. As of June 2026, frontier-tier models cost roughly low single-dollars per million input tokens and several times that per million output tokens. Prompt caching cuts repeated input costs by up to 90 percent; batch processing cuts asynchronous workload costs by roughly half. Confirm current rates on Anthropic's pricing page before budgeting.

Pricing references are as of June 2026 and may change. Always confirm current pricing on the vendor site before committing.

The pricing tiers

Claude offers several model tiers, each priced differently. Pricing changes; treat these as relative ranges as of June 2026 and confirm on Anthropic's site:

Frontier reasoning tier (Opus or equivalent). Highest input and output prices per million tokens. Best for hardest reasoning, longest output, highest-stakes work.
General-purpose tier (Sonnet or equivalent). Lower per-token prices. Best for the bulk of production workloads. Most teams default here.
Fastest tier (Haiku or equivalent). Cheapest per token. Best for high-volume, low-stakes work and classification.

What drives the number up or down

Model tier. The largest single lever. Pick the cheapest model that handles the work well.
Input vs output split. Output is several times more expensive than input. Long prompts with short answers are cheap; short prompts with long answers are expensive.
Prompt caching. Cuts repeated-input costs by up to 90 percent. Huge win for assistants with stable system prompts or loaded documents.
Batch processing. Asynchronous batch API cuts costs roughly in half for workloads that can wait.
Context window usage. Larger contexts cost more on each call; only pass what the model needs.

Where it fits the cost ladder

Option	Typical cost	Best when
Claude API (per use)	Low single-dollars per million input tokens at top tier; several times that per million output	Custom workflows, agents, internal tools
Claude Pro (subscription)	~$20 / month	Individual interactive use with no API integration
Claude Team	Per-seat published rate	Small orgs wanting shared workspace
Claude Enterprise	$60 to $200-plus / seat / month custom	Larger orgs needing SSO, audit, custom terms
Standalone AI products	$200 to $2,000 / month	Pre-built workflows on top of the API

Who should pay for it

Use the API when you are building an internal tool, an agent, or an integration into your CRM or product.
Use Pro or Team when your use is interactive humans working in a chat surface.
Cache aggressively when you have a stable system prompt or loaded documents that get reused across many calls.
Batch when the workload can tolerate a few hours of delay (classification, enrichment, summarization at scale).

Frequently asked questions

How is the Claude API priced?

Per million tokens of input and per million tokens of output, separately, with different rates per model tier. Output is several times more expensive than input.

How much does a typical chatbot conversation cost?

Highly variable. A conversation with a few thousand input tokens and an output of a few hundred tokens generally costs cents on Sonnet tier. With prompt caching the realized cost drops substantially.

Should I cache?

Almost always, if you have any reused prompt content. Cached input tokens are up to 90 percent cheaper. The setup is minor.

Should I batch?

If your workload is asynchronous (overnight enrichment, classification, summarization), yes. Batch API discounts cut costs by roughly half.

How current is this estimate?

As of June 2026. Anthropic adjusts API pricing periodically; confirm on the pricing page before budgeting.

Designing your AI stack?

The free AI Tool Stack Auditor surfaces redundancies and gaps in 3 minutes. The $1,500 AI Audit goes deeper: a written roadmap in 5 business days.

Free auditor →Book AI Audit →

Next step

Want this priced against your actual stack?

The $1,500 Treetop AI Audit produces a written 5-business-day budget for your team, your industry, and your goals.

✓ Money-back guarantee. If the AI Audit does not surface 10x its $1,500 cost in savings or revenue, you get a refund. Real outcome: How a finance firm deployed Claude across 4 functions →

Book the AI Audit →