How much does Claude Opus 4.7 cost per 1M tokens?

Claude Opus 4.7 costs $5.00 per 1M input tokens and $25.00 per 1M output tokens. The rate card matches Opus 4.6 exactly, but the retrained tokenizer emits ~15% more tokens for the same text - so invoices typically run 15% higher than Opus 4.6 for identical work.

What is the cheapest Anthropic model?

Claude Haiku 4.5 is the cheapest Claude model at $1.00 input / $5.00 output per 1M tokens. Cheap enough to run as a background content moderator, classifier, or high-volume summariser.

Does Anthropic offer batch pricing discounts?

Yes. The Message Batches API offers a 50% discount on both input and output tokens in exchange for up to 24-hour completion. Available for every Claude model. Combine with prompt caching for compounding savings on RAG and document-processing workloads.

How does Anthropic pricing compare to competitors?

Claude Opus 4.7 is Anthropic's flagship at $5/$25 per 1M tokens. That is more expensive than OpenAI GPT-5.4 ($2.50/$15) and Google Gemini 2.5 Pro ($1.25/$10) at the frontier tier. Anthropic's advantage is the 1M context window on Opus 4.6/4.7 and Sonnet 4.6, plus the extended-thinking mode that often needs fewer retries.

What is the Anthropic free tier?

Anthropic offers a limited free tier on the Claude consumer product but the API itself is pay-as-you-go with no free tokens. New accounts typically get a small credit for experimentation. The countTokens endpoint (used by Calcis to exactly count input tokens for Claude models) is itself free - we pay nothing to tell you what a prompt costs.

Anthropic

Anthropic API Pricing (2026)

The Claude family from Haiku to Opus, with prompt caching, batch discounts, and extended thinking explained. Every rate computed at today's prices, every ratio live from the estimator.

Pricing verified 2026-04-17

Complete rate card

Every Anthropic model Calcis tracks, sorted cheapest to most expensive on a typical chat-shape request (1k in, 2k out). Rates are USD per 1M tokens.

Model	Input / 1M	Output / 1M	Context	Max out	Details
Claude Haiku 4.5	$1.00	$5.00	200K	64K	View →
Claude Sonnet 4.6	$3.00	$15.00	1M	64K	View →
Claude Sonnet 4.5	$3.00	$15.00	200K	64K	View →
Claude Opus 4.6	$5.00	$25.00	1M	128K	View →
Claude Opus 4.5	$5.00	$25.00	200K	64K	View →
Claude Opus 4.7	$5.00	$25.00	1M	128K	View →
Claude Opus 4.1	$15.00	$75.00	200K	32K	View →

Every Anthropic model

Each page has a headline price card, cost ladder, a narrative on when to reach for that model, and an FAQ with the common pricing questions answered from the live rate card.

Claude Haiku 4.5

200K ctx

$1.00 in · $5.00 out · per 1M tok

Claude Sonnet 4.6

1M ctx

$3.00 in · $15.00 out · per 1M tok

Claude Sonnet 4.5

200K ctx

$3.00 in · $15.00 out · per 1M tok

Claude Opus 4.6

1M ctx

$5.00 in · $25.00 out · per 1M tok

Claude Opus 4.5

200K ctx

$5.00 in · $25.00 out · per 1M tok

Claude Opus 4.7

1M ctx

$5.00 in · $25.00 out · per 1M tok

Claude Opus 4.1

200K ctx

$15.00 in · $75.00 out · per 1M tok

What does Claude Haiku 4.5 vs Claude Opus 4.1 actually cost?

Four workload shapes from tiny to massive. Output is roughly half the input (a typical chat/completion pattern). The ratio column shows the Anthropic spread at a glance.

Scenario	Tokens (in / out)	Claude Haiku 4.5	Claude Opus 4.1	Ratio
Tiny	100 / 50	$0.00035	$0.00525	15.0×
Short request	1,000 / 500	$0.00350	$0.0525	15.0×
Long document	10,000 / 5,000	$0.0350	$0.5250	15.0×
Massive context	100,000 / 50,000	$0.3500	$5.25	15.0×

Discounts & modifiers

Prompt caching. Anthropic charges separate rates for cache reads (cheap) and cache writes (expensive - typically 25% more than base input). The caching sweet spot is a long, static system prompt followed by short per-request user turns: write the cache once, read it on every subsequent call at ~10% of the input rate. Claude Sonnet 4.6 and Opus 4.6/4.7 publish these rates; Haiku 4.5 does not advertise a cache-read discount on its current rate card.

Batch API. 50% discount on both input and output in exchange for 24-hour turnaround, available on every Claude model. Combine with prompt caching for compounding savings: a RAG pipeline with a cached corpus running through batch can bill at ~5% of the base rate card.

Opus 4.7 tokenizer rebill. The rate card is identical to Opus 4.6 ($5/$25 per 1M) but the retrained tokenizer emits 1.0-1.35× more tokens for the same text, mean ~15% more. Calcis bakes the 1.15× midpoint into every Opus 4.7 estimate so the forecast matches the invoice instead of surprising you after the fact.

Extended thinking.Claude Opus 4.7 and Sonnet 4.6 support extended thinking (a thinking budget parameter). Thinking tokens bill at the output rate. Calcis' estimator exposes a thinking-effort selector so you can see how each level (low / medium / high / extra-high / max) changes cost before you send.

Which Anthropic model should I use?

For simple tasks and high-volume workloads, Claude Haiku 4.5 is the right default. At $1/1M input and $5/1M output it is ~5x cheaper than Sonnet and ~25x cheaper than Opus, and the 200K context window is enough for any single-document task.

For production chat, RAG, and coding assistants, Claude Sonnet 4.6 is the workhorse. The 1M context window (matched by Opus 4.6/4.7) is rare at this price point - you can stuff full codebases or multi-document research into a single request without chunking.

For the hardest reasoning (ambiguous specs, multi-hour autonomous agents, research-grade synthesis), Claude Opus 4.7 is the flagship. Extended thinking costs more output tokens, but for tasks where a cheaper model loops and retries, Opus often comes out cheaper in total spend.

Claude Opus 4.7 ships with a retrained tokenizer that emits ~1.15x more tokens than Opus 4.6 for the same text. Calcis bakes this multiplier into every Opus 4.7 estimate so the forecast matches the invoice.

Estimate your Anthropic costs →

Drop a prompt into the estimator, pick any Anthropic model, and get the exact dollar cost - input tokens counted with Anthropic's own tokenizer, output tokens predicted by our regression model.

Open the estimator →Compare side-by-side →

Frequently asked

How much does Claude Opus 4.7 cost per 1M tokens?: Claude Opus 4.7 costs $5.00 per 1M input tokens and $25.00 per 1M output tokens. The rate card matches Opus 4.6 exactly, but the retrained tokenizer emits ~15% more tokens for the same text - so invoices typically run 15% higher than Opus 4.6 for identical work.
What is the cheapest Anthropic model?: Claude Haiku 4.5 is the cheapest Claude model at $1.00 input / $5.00 output per 1M tokens. Cheap enough to run as a background content moderator, classifier, or high-volume summariser.
Does Anthropic offer batch pricing discounts?: Yes. The Message Batches API offers a 50% discount on both input and output tokens in exchange for up to 24-hour completion. Available for every Claude model. Combine with prompt caching for compounding savings on RAG and document-processing workloads.
How does Anthropic pricing compare to competitors?: Claude Opus 4.7 is Anthropic's flagship at $5/$25 per 1M tokens. That is more expensive than OpenAI GPT-5.4 ($2.50/$15) and Google Gemini 2.5 Pro ($1.25/$10) at the frontier tier. Anthropic's advantage is the 1M context window on Opus 4.6/4.7 and Sonnet 4.6, plus the extended-thinking mode that often needs fewer retries.
What is the Anthropic free tier?: Anthropic offers a limited free tier on the Claude consumer product but the API itself is pay-as-you-go with no free tokens. New accounts typically get a small credit for experimentation. The countTokens endpoint (used by Calcis to exactly count input tokens for Claude models) is itself free - we pay nothing to tell you what a prompt costs.

Pricing verified 2026-04-17 from Anthropic's published rate card.