VynarisEarly betaGet my API key

Blog · 2026-10-09 · By , Founder

Claude API Pricing: Every Model, Cache and Batch Rate

Claude API pricing verified 2026-10-09: Haiku 5.5 $0.10/$0.50, Sonnet 5.5 $2/$10, Opus 5.5 $4/$20, Fable 5.1 $10/$50 per MTok, plus caching, batch, long-context and cost per task.

TL;DR: Claude API pricing on 2026-10-09 runs from $0.10 input / $0.50 output per million tokens (Claude Haiku 5.5) to $10 / $50 (Claude Fable 5.1). Claude Sonnet 5.5 costs $2 / $10 and Claude Opus 5.5 costs $4 / $20. Batch halves every rate, and cache hits cost 2.5% to 10% of the input price.

Prices verified 2026-10-09 against the Anthropic pricing docs. Anthropic bills the Claude API per million tokens (MTok), with separate input and output rates. On a representative task of 10,000 input tokens and 1,000 output tokens, that works out to $0.0015 on Haiku 5.5, $0.03 on Sonnet 5.5, $0.06 on Opus 5.5 and $0.15 on Fable 5.1. The rest of this page covers the full rate card, the modifiers that move a real bill (caching, batch, long context, data residency, fast mode, tools) and cost per task with every assumption shown.

How much does the Claude API cost per million tokens?

Claude API price is set per model, per million tokens, in USD. These are the current models as listed on the Anthropic docs on 2026-10-09:

Model                                 Input / MTok  Output / MTok  5m cache write  1h cache write  Cache hit
------------------------------------  ------------  -------------  --------------  --------------  ---------
Claude Fable 5.1                      $10           $50            $12.50          $20             $0.25
Claude Opus 5.5                       $4            $20            $5              $8              $0.20
Claude Sonnet 5.5                     $2            $10            $2.50           $4              $0.10
Claude Haiku 5.5 (prompt up to 100k)  $0.10         $0.50          $0.125          $0.20           $0.01
Claude Haiku 5.5 (prompt over 100k)   $0.50         $2.50          $0.625          $1              $0.05

Output costs 5x input on every model. For most chat and agent workloads, output is the line that dominates the bill even though there are fewer output tokens.

What do older Claude models cost?

Older Claude models stay on the API at their own rates. If you still call them, check whether a newer model is cheaper:

Model                                     Input / MTok  Output / MTok  Cache hit
----------------------------------------  ------------  -------------  ---------
Claude Opus 5, Opus 4.8, Opus 4.5 to 4.7  $5            $25            $0.50
Claude Opus 4.1, Opus 4                   $15           $75            $1.50
Claude Sonnet 5                           $2            $10            $0.20
Claude Sonnet 4.6, 4.5, 4                 $3            $15            $0.30
Claude Haiku 4.5                          $1            $5             $0.10
Claude Haiku 3.5                          $0.80         $4             $0.08

Opus 5.5 is 20% cheaper than Opus 5 on both input and output, and its cache hit is 60% cheaper ($0.20 vs $0.50). Haiku 5.5 is 90% cheaper than Haiku 4.5 below the 100k-token threshold. Anyone still on Opus 4.1 pays 3.75x the Opus 5.5 rate.

Does the newer tokenizer change the real price?

Yes. Anthropic states that Claude 4.7 and later models use a tokenizer that produces about 30% more tokens for the same text. Sonnet 4.6 and earlier use the old one. So list prices across that line are not directly comparable.

Worked example, same text: Sonnet 4.6 at $3 input. Sonnet 5.5 at $2 input times 1.3 tokens is an effective $2.60, so it is still about 13% cheaper, not the 33% the sticker suggests. On output, $10 x 1.3 = $13 vs $15, also 13% cheaper. We measured the tokenizer effect in more detail in the Anthropic tokenizer post.

How does prompt caching pricing work on Claude?

Prompt caching charges a premium to write a prompt prefix into cache and a deep discount to read it back. The multipliers relative to base input price:

A 5-minute write pays for itself after one cache read. A 1-hour write pays for itself after two reads. Caching is enabled either with a top-level cache_control field (automatic) or with cache_control on individual content blocks. The multipliers stack with batch and data residency. Our prompt caching break-even analysis covers when the 1-hour TTL is worth it.

How much does the Batch API save?

The Message Batches API gives a flat 50% discount on input and output for asynchronous jobs. Batch rates: Fable 5.1 $5 / $25, Opus 5.5 $2 / $10, Sonnet 5.5 $1 / $5, Haiku 5.5 $0.05 / $0.25 (up to 100k prompt). Fast mode cannot be combined with batch. See batch APIs compared for the OpenAI side.

What does a Claude API task actually cost?

Cost per task is the number that matters for a budget. Our assumptions, which you can change:

Formula: cost = input_tokens x input_rate / 1,000,000 + output_tokens x output_rate / 1,000,000.

Model              Uncached / 1,000 tasks  Cached / 1,000 tasks  Batch, uncached / 1,000 tasks
-----------------  ----------------------  --------------------  -----------------------------
Claude Fable 5.1   $150                    $72                   $75
Claude Opus 5.5    $60                     $29.60                $30
Claude Sonnet 5.5  $30                     $14.80                $15
Claude Haiku 5.5   $1.50                   $0.78                 $0.75

Derivation for Sonnet 5.5 cached: 2,000 x $2 / 1M = $0.004, plus 8,000 x $0.10 / 1M = $0.0008, plus 1,000 x $10 / 1M = $0.01, total $0.0148 per task. Plug your own token counts into the LLM cost calculator.

The gap between Haiku 5.5 and Fable 5.1 is 100x on this workload. That spread is the reason routing matters: classification and extraction rarely need Fable.

What is the Haiku 5.5 long-context price cliff?

Claude Haiku 5.5 is priced by prompt length. A request whose prompt is over 100,000 tokens pays $0.50 / $2.50 instead of $0.10 / $0.50, and the prompt length counts cache reads and writes too. Each request is priced on its own.

Example: a 150,000-token prompt with 1,000 output tokens costs 150,000 x $0.50 / 1M + 1,000 x $2.50 / 1M = $0.0775. Split into two 75,000-token requests it costs 2 x (75,000 x $0.10 / 1M + 1,000 x $0.50 / 1M) = $0.016, about 79% less, if your task can be chunked. All other Claude 4.6+ models bill the full 1M context window at standard rates.

What other charges appear on a Claude API bill?

Several line items sit outside the token rates:

Is a Claude subscription cheaper than the API?

A Claude subscription is a flat fee for the chat apps and Claude Code: Pro is $20 per month ($17 billed annually) and Max starts at $100 per month, per the Claude pricing page. Subscriptions carry usage limits and are not usable as API credits for your own app. For heavy Claude Code users the break-even is real; we did the math in Claude Max vs API pay-per-token.

When is the expensive model worth it?

Fable 5.1 at $50 per million output tokens is worth it when a cheaper model needs several retries or extra turns to finish. If Sonnet 5.5 needs more than 5x the tokens to finish a task, Fable 5.1 is cheaper per completed task. That is common on long-horizon agent work and rare on extraction or classification. We tested this in Fable 5.1 vs Opus 5 token efficiency.

Where Vynaris fits

Vynaris is an OpenAI-compatible router that sends each request to the cheapest model that can handle it, including the Claude models above. If most of your traffic is simple and a minority needs Opus or Fable, routing captures the 20x to 100x spread in the table above. If every request truly needs Fable 5.1, a router saves you nothing on Claude itself; call Anthropic directly.

FAQ

How much does Claude cost per 1,000 tokens?

Divide the per-million rate by 1,000. Sonnet 5.5 costs $0.002 per 1,000 input tokens and $0.01 per 1,000 output tokens.

What is the cheapest Claude model?

Claude Haiku 5.5 at $0.10 input / $0.50 output per MTok for prompts up to 100,000 tokens, or $0.05 / $0.25 through the Batch API.

Is the Claude API free?

No. The Claude API is pay-as-you-go with prepaid credits. The free tier on claude.ai covers the chat app, not API calls.

Does Claude charge for input tokens?

Yes. Input, output, cache writes and cache reads are each billed at their own rate, and tool schemas count as input tokens.

Is Claude API pricing the same on AWS Bedrock?

Global endpoint pricing matches Anthropic's list, but Bedrock and Google Cloud set their own prices and charge a 10% premium for regional endpoints. Check the Amazon Bedrock pricing page.

How do I estimate my Claude API bill?

Count input and output tokens per task, multiply by the rates above, then multiply by monthly tasks. The token usage fields in each API response give you exact counts.

Sources