Claude API Pricing: Haiku 4.5, Sonnet 4.6, and Opus 4.6 (April 2026)

Anthropic's public Claude pricing is much simpler than many comparison pages make it look. The current public model pages expose three standard API tiers: Haiku 4.5, Sonnet 4.6, and Opus 4.6. Those are the rows that matter for most teams building on Claude today.

This guide uses Anthropic's current public model pages for Haiku 4.5, Sonnet 4.6, and Opus 4.6.

Claude pricing at a glance

Model	Input $/M	Output $/M	Notes
Claude Haiku 4.5	$1.00	$5.00	Fastest, cheapest Claude tier; also available in Claude Code
Claude Sonnet 4.6	$3.00	$15.00	Default production tier; 1M context beta on API only
Claude Opus 4.6	$5.00	$25.00	Premium tier; 1M context beta on Claude Platform only

Every current public Claude tier still keeps the same 5x output-to-input ratio, which makes back-of-the-envelope budgeting easy.

What Anthropic says about discounts

Anthropic repeats the same two cost levers across the current model pages:

• Prompt caching: up to 90% cost savings
• Batch processing: 50% cost savings

If you reuse long system prompts, stable few-shot prefixes, or repeated document headers, prompt caching is usually the fastest way to shrink your bill. If your workload is asynchronous, Batch is the bigger blunt instrument.

Which Claude tier fits which job

Haiku 4.5

Anthropic positions Haiku as the fast, cost-efficient model for scaled deployments, real-time applications, and coding sub-agents. If your workload is latency-sensitive or cost-sensitive, this is the first Claude model to try.

Sonnet 4.6

Sonnet 4.6 is the default production tier. It is the current middle ground for coding, agents, and professional workflows, and Anthropic says the 1M token context window is available in beta on the API only.

Opus 4.6

Opus 4.6 is no longer priced as an extreme outlier relative to Sonnet. It is still the premium tier, but the published public price is $5/$25, not the much higher numbers that older comparison tables often repeat. Anthropic also says US-only inference is available at 1.1x pricing for Opus 4.6.

Real cost example: the same workload on all three Claude tiers

Assume a RAG workflow serving 3,000 requests per day, each with 2,400 input tokens and 350 output tokens.

Haiku 4.5

• Daily input: 3,000 x 2,400 x $1.00 / 1M = $7.20
• Daily output: 3,000 x 350 x $5.00 / 1M = $5.25
• Monthly total: about $373.50

Sonnet 4.6

• Daily input: 3,000 x 2,400 x $3.00 / 1M = $21.60
• Daily output: 3,000 x 350 x $15.00 / 1M = $15.75
• Monthly total: about $1,120.50

Opus 4.6

• Daily input: 3,000 x 2,400 x $5.00 / 1M = $36.00
• Daily output: 3,000 x 350 x $25.00 / 1M = $26.25
• Monthly total: about $1,867.50

That is the pricing shape to remember:

• Haiku is roughly one-third the cost of Sonnet
• Opus is roughly 1.7x Sonnet
• The jump from Sonnet to Opus is meaningful, but not remotely as dramatic as some stale 2025-era tables suggest

The practical takeaway

If you want a default:

• Use Haiku 4.5 for high-volume pipelines, routing, and latency-sensitive UX.
• Use Sonnet 4.6 as the standard production Claude tier.
• Use Opus 4.6 when a measured quality gain justifies the premium.

And before you obsess over prompt wording, turn on the two discounts Anthropic is already telling you to use: prompt caching and batch processing.

For a broader vendor comparison, see our LLM pricing overview.