How much does Claude Haiku 5.5 cost?

$0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Over 100,000 tokens, $0.50 input and $2.50 output. Cache reads are $0.01 under the line and $0.05 over. Batch is 50% off.

How does Haiku 5.5 compare with GPT-6 Luna?

Same short-prompt rates: $0.10 input and $0.50 output per million tokens. Luna's short context runs to 272,000 input tokens and its long-context rates are $0.20 and $0.75. Haiku 5.5 switches to 5x rates over 100,000 tokens. The two models count tokens differently, so test on your own text.

What does a million lead decisions cost on Haiku 5.5?

At 1,000 input tokens and 50 output tokens per decision, about $125 at list, or $0.125 per 1,000 decisions. Haiku 4.5 costs $1,250 for the same token counts. Batch cuts Haiku 5.5 to $62.50.

What happens over 100,000 tokens?

Every Haiku 5.5 rate rises 5x. A brief with 101,000 input tokens and 2,000 output tokens costs $0.0555, against $0.0109 at 99,000 input tokens. Anthropic has not published whether cached tokens count toward the line.

Do Claude Team plans include API credits?

Yes, from the week of Oct 7 2026. $20 per Standard seat and $100 per Premium seat each month, pooled and capped at $500. Max 5x gets $100 and Max 20x gets $200. Free, Pro, and Enterprise are not eligible. New subscribers wait 7 days.

Do unused Claude API credits roll over?

No. They expire at the end of each billing cycle, or monthly on annual plans. If the organization has no purchased credits, API requests stop when the credits run out.

What should go on the Order Form?

A price lock on the under-100,000 rates, a written threshold definition, whole-request or overage billing, a prompt size cap, a tokenizer test, credits treated as pilot money, credit scope by cloud, workspace spend caps, and the model retirement date.

← Reality Check

Claude Haiku 5.5 Costs $0.10 per Million Tokens Under 100K and 5x More Over It. Where Do Your Lead Prompts Land?

Anthropic released Claude Haiku 5.5 on Oct 7 2026 at $0.10 per million input tokens and $0.50 per million output for prompts up to 100,000 tokens, the same as OpenAI's GPT-6 Luna. Over 100,000 tokens every rate jumps 5x. A million short lead decisions costs about $125, against $1,250 on Haiku 4.5. Team plans now pool up to $500 a month in API credits that expire every cycle. Here is the math and the Order Form lines.

Jonathan Kvarfordt · Published October 7, 2026 · 7 min read

Why trust this analysis?

The short answer

Claude Haiku 5.5 lists $0.10 per million input tokens under 100,000 tokens and 5x more over it. What does a high-volume lead classification workload cost, do Team plan API credits cover a pilot, and what goes on the Order Form?

Anthropic's pricing docs list Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, and $0.50 and $2.50 for prompts over 100,000. Cache reads are $0.01 under the line and $0.05 over it. Batch is 50% off. Haiku 4.5 lists $1 and $5. OpenAI lists GPT-6 Luna at $0.10 and $0.50 in short context, which runs to 272,000 input tokens. A lead decision with 1,000 input tokens and 50 output tokens costs $0.000125 on Haiku 5.5 or Luna, about $125 per million, against $1,250 per million on Haiku 4.5. Haiku 5.5 counts the same text as about 30% more tokens than Haiku 4.5, so test on your own records. The cliff hits long prompts: an account brief at 101,000 input tokens costs 5.1x the same brief at 99,000. Team plans earn $20 per Standard seat and $100 per Premium seat in monthly API credits, pooled and capped at $500. A pilot of 50,000 lead decisions and 20,000 short call summaries costs about $31 a month at Haiku 5.5 rates, inside even a two-seat Team pool. Credits expire each cycle with no rollover, work only on the Claude Console, and API requests stop at zero if no paid balance exists. On the Order Form: a price lock, a written threshold definition, a prompt size cap, a tokenizer test, workspace spend caps, and the model retirement date.

Decision rule

Run lead decisions on Haiku 5.5 only with a hard prompt cap below 100,000 tokens and a written answer on which tokens count toward the line.

Evidence

Claude Platform docs, Pricing (retrieved Oct 7 2026): Claude Haiku 5.5 $0.10 input, $0.50 output, $0.125 5-minute cache write, $0.20 1-hour cache write, $0.01 cache read for prompts up to 100,000 tokens; $0.50, $2.50, $0.625, $1, $0.05 for prompts over 100,000. Batch $0.05 and $0.25 under, $0.25 and $1.25 over. "Claude Haiku 5.5 is priced by prompt length: a prompt of over 100,000 tokens pays higher prices."
Anthropic, Introducing Claude Haiku 5.5 (Oct 7 2026): about 75% lower cost to run than Haiku 4.5 on average; 90% lower under 100,000 tokens and 50% lower over; about 90% of Haiku 4.5 requests were under 100,000 tokens. Sonnet 5.5 cache reads cut from $0.20 to $0.10. Monthly API credit: Max 5x $100, Max 20x $200, Team up to $500 pooled.
Claude Platform docs, Claude Haiku 5.5 overview: same text counts as approximately 30% more tokens than on Claude Haiku 4.5; retirement not sooner than October 7, 2027.
OpenAI API pricing (retrieved Oct 7 2026): gpt-6-luna $0.10 input, $0.01 cached input, $0.50 output in short context; $0.20, $0.02, $0.75 in long context. Short context is 272K input tokens or fewer.
Claude Platform docs and Help Center, API credits for Max and Team plans: $20 per Standard seat and $100 per Premium seat, pooled, capped at $500; no rollover; 7-day wait for new subscribers; Free, Pro, and Enterprise not eligible; Claude Console only.

Operator action

This week: pull token counts for your lead routing, scoring, and summary prompts on Haiku 5.5. Count how many cross 100,000 tokens. Put a hard cap in code, fence the pilot in its own workspace with a spend limit, and send Anthropic the threshold questions before you sign.

Last reviewed October 7, 2026

Anthropic just priced a small model like a rounding error. Then it put a 5x cliff at 100,000 tokens.

On Oct 7 2026 Anthropic released Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens (Anthropic launch). Over 100,000 tokens, the same model costs $0.50 input and $2.50 output (Claude pricing docs). Haiku 4.5 lists $1 and $5. OpenAI lists GPT-6 Luna at $0.10 and $0.50 in short context (OpenAI pricing).

The argument

How this reality check breaks down

A map of the sections ahead, in the order the case is made. The sourced charts are in the research library.

Anthropic says Haiku 5.5 costs about 75% less to run than Haiku 4.5 on average. Its footnote shows the math: 90% lower under 100,000 tokens, 50% lower over it, about 90% of Haiku 4.5 requests under the line, and a new tokenizer that uses slightly more tokens per task. That 75% is a vendor average. Your number depends on where your prompts land.

The rate card

Up to 100,000 tokens. Input $0.10, output $0.50, cache read $0.01, 5-minute cache write $0.125, 1-hour cache write $0.20, per million tokens (Claude pricing docs).
Over 100,000 tokens. Input $0.50, output $2.50, cache read $0.05, 5-minute cache write $0.625, 1-hour cache write $1. Every rate is 5x.
Batch. 50% off input and output: $0.05 and $0.25 under the line, $0.25 and $1.25 over it.
US-only inference. 1.1x on input and output tokens (Anthropic pricing).
Model life. Model ID claude-haiku-5-5, 1M token context, 128K max output. Retirement not sooner than October 7, 2027 (Haiku 5.5 overview).

Reality Check

Claude Haiku 5.5 Costs $0.10 per Million Tokens Under 100K and 5x More Over It. Where Do Your Lead Prompts Land?

Source-cited charts live in the research library.

Where the line sits

Anthropic's docs say it in one line: "a prompt of over 100,000 tokens pays higher prices." Claude 4.6 and later models bill a 900,000-token request at the same per-token rate as a 9,000-token request. Haiku 5.5 is the listed exception (Claude pricing docs).

GPT-6 Luna draws its line at 272,000 input tokens. Over it, Luna goes to $0.20 input and $0.75 output, 2x and 1.5x (OpenAI pricing). Haiku 5.5 draws its line lower and jumps 5x on every rate.

Anthropic has not published three things: whether cached tokens count toward the 100,000, how a request of exactly 100,000 tokens is billed, and in plain words whether the higher rate covers the whole request. VentureBeat flagged the first two in its launch coverage (VentureBeat). The docs read as the whole prompt moving to the higher rate. Anthropic's caching docs add up cache reads, cache writes, and uncached input as total input tokens processed (prompt caching docs). If the 100,000 test uses that total, a large cached system prompt counts. Get the answer in writing.

What 1,000 lead decisions cost

Take one lead routing or scoring call: 1,000 input tokens for the record and the rules, 50 output tokens for a label and a reason. Same shape as our OpenAI Decisions API decoder.

Haiku 5.5, under 100,000. 1,000 x $0.10 + 50 x $0.50, per million tokens = $0.000125 a decision. $0.125 per 1,000 decisions. $125 per million.
GPT-6 Luna, short context. Same rates. $0.125 per 1,000. $125 per million.
Haiku 4.5. 1,000 x $1 + 50 x $5 = $0.00125 a decision. $1.25 per 1,000. $1,250 per million.
Haiku 5.5 on Batch. $0.0625 per 1,000. $62.50 per million, if routing can wait.

One catch. Anthropic says Haiku 5.5 counts the same text as about 30% more tokens than Haiku 4.5 (Haiku 5.5 overview). A lead record that runs 1,000 tokens on Haiku 4.5 runs about 1,300 on Haiku 5.5. That puts Haiku 5.5 near $162.50 per million decisions, still about 87% below Haiku 4.5. OpenAI uses its own tokenizer. Count your own records on both models before you call Haiku and Luna a tie.

Two bar charts. Panel A, one million short lead decisions at 1,000 input and 50 output tokens: Claude Haiku 4.5 $1,250, Claude Haiku 5.5 $125, OpenAI GPT-6 Luna $125, Claude Haiku 5.5 Batch $62.50. Panel B, 1,000 long account briefs with 2,000 output tokens: Haiku 5.5 at 99K input $10.90, Haiku 5.5 at 101K input $55.50, GPT-6 Luna at 101K input $11.10, Haiku 4.5 at 101K input $111. — List cost at standard API rates, Oct 7 2026. Sources: Anthropic pricing page, Claude Platform pricing docs, Haiku 5.5 launch post; OpenAI API pricing. Token counts as billed per model. Our arithmetic.

Where the cliff bites RevOps

A lead decision sits far below 100,000 tokens. Account briefs can cross it. Those are the prompts that stuff call transcripts, email threads, and CRM history into one request.

At 99,000 input tokens and 2,000 output tokens, Haiku 5.5 costs $0.0109 a brief, or $10.90 per 1,000. At 101,000 input tokens it costs $0.0555, or $55.50 per 1,000. Two thousand more input tokens, 5.1x the bill. GPT-6 Luna at 101,000 input tokens stays at $11.10 per 1,000, because its line sits at 272,000. Haiku 4.5 costs $111 per 1,000 at that size.

The fix lives in your code. Cap prompt size below 100,000 tokens. Summarize or split long histories first. Log every request that crosses the line, with the record ID.

Do Team plan API credits cover a RevOps pilot?

Anthropic is rolling out monthly Claude Platform API credits to Max and Team subscribers this week (Anthropic launch). Max 5x gets $100 a month. Max 20x gets $200. Team plans get $20 per Standard seat and $100 per Premium seat, pooled into one balance capped at $500 (API credits docs). Anthropic's own example: three Standard seats and two Premium seats get $260 a month. Team Standard seats list at $20 a month billed annually or $25 monthly. Premium seats list at $100 or $125 (Anthropic pricing).

Price a small pilot at Haiku 5.5 rates. 50,000 inbound lead decisions a month at 1,000 input and 50 output tokens: $6.25. 20,000 call summaries at 10,000 input and 500 output tokens: $25. Total: $31.25 a month. Team plans start at two seats, and two Standard seats earn $40 a month in credits. The $500 cap covers about 4 million short lead decisions at Haiku 5.5 rates, or 400,000 at Haiku 4.5 rates.

Now push those 20,000 summaries over 100,000 input tokens each. The summaries alone cost $1,035 a month. That is past the $500 cap, and the pilot stops.

Credit fine print

No rollover. Unused credits expire at the end of each billing cycle, or monthly on annual plans (Help Center).
Requests stop at zero. If the linked organization has no purchased credits, API requests stop until next month's credits arrive. Usage is never charged to your Claude plan. Organizations invoiced through Anthropic sales are billed as usual past the credits (API credits docs). A lead router that stops mid-month is a pipeline problem.
Console only. Credits cover the Claude API, Message Batches, Claude Managed Agents, the Claude Agent SDK, and the Console playground. They do not cover Claude Code, extra usage in the Claude apps, Amazon Bedrock, Google Cloud Vertex AI, or Microsoft Foundry.
Who is out. Free, Pro, and Enterprise plans are not eligible. New subscribers wait 7 days. Claim on claude.ai in a web browser.
One organization. Each plan links to one Console organization, and you cannot change the link yourself. Every API key in that organization draws from the same balance. Fence a project with a workspace spend limit.
Seat timing. The pool follows the seats on your plan at the start of each billing month. Seats you add count from the next month.
Terms. Linking accepts Anthropic's Supplemental Credit Terms. Credits are non-refundable, cannot be transferred, and have no cash value (credit terms).
Usage tier. Linked organizations move to at least the Start tier, and to at least the Build tier if monthly credits exceed $200. Credits do not count toward higher tiers.

The Sonnet 5.5 cache cut

Same day, Anthropic halved Sonnet 5.5 cache reads from $0.20 to $0.10 per million tokens. Anthropic says that cuts Sonnet 5.5 cost on most agentic work by about 20% (Anthropic launch; Unite.AI). Input and output stay at $2 and $10. Our Sonnet 5.5 decoder used the $0.20 launch rate. Use $0.10 from Oct 7 2026. If your agents re-read a large cached prompt every turn, check the cache-read line on your next invoice.

For seat math on Claude subscriptions, see Claude's $200 plan vs ChatGPT Pro. For hard caps on API keys, see OpenAI usage tiers and the spend limit.

Order Form lines for Claude Haiku 5.5

Price lock. Hold $0.10 input and $0.50 output under 100,000 tokens for the term. Anthropic's pricing page says prices and plans can change at its discretion.
Threshold definition. Get in writing which tokens count toward 100,000: cache reads, cache writes, tool definitions, system prompt, images. Per request, measured how.
Whole request or overage. Confirm whether the over-100,000 rate applies to every token in the request, and how a request of exactly 100,000 tokens bills.
Prompt size cap. A hard cap below 100,000 tokens in your code, plus an alert and a log line on every request that crosses.
Tokenizer test. Count tokens on 500 of your own records on Haiku 4.5, Haiku 5.5, and any rival model. Price from those counts.
Credits are pilot money. No rollover, monthly expiry, and requests stop at zero. Production routing runs on a purchased balance or an invoiced account.
Credit scope. Console only. If you buy Claude through Bedrock, Vertex AI, or Foundry, or you are on Enterprise, the credits do not apply. Ask for an equal pilot credit in the deal.
Spend cap per workspace. Put the router key in its own workspace with a spend limit. Auto-reload off, or capped.
Model life. Anthropic lists retirement not sooner than October 7, 2027. Tie your rate and notice period to that date.

Who owns which line

Procurement: the price lock, the threshold definition in writing, and the model-life date.
RevOps and GTM engineering: the prompt size cap, the tokenizer test, and the crossing log.
Revenue Finance: workspace spend caps, credit expiry, and auto-reload settings.
IT and platform: which cloud runs the model. Credits only apply on the Claude Console.

Soft opens

The 75% saving is Anthropic's average across Haiku 4.5 traffic. Benchmarks and customer quotes on the launch page are vendor-reported.
Anthropic has not published whether cached tokens count toward the 100,000 line or how exactly 100,000 bills.
Anthropic says credits roll out over a few days. Some eligible accounts will not see them yet.
All prices are list. Committed-spend deals may price differently.
GPT-6 Luna and Haiku 5.5 count tokens differently. Equal rates do not guarantee equal bills on the same text.

Sources

Anthropic, Introducing Claude Haiku 5.5 (Oct 7 2026): anthropic.com/claude-haiku-5-5
Claude Platform docs, Pricing (retrieved Oct 7 2026): platform.claude.com
Anthropic, Pricing page (retrieved Oct 7 2026): anthropic.com/pricing
Claude Platform docs, Claude Haiku 5.5 overview: platform.claude.com
Claude Platform docs, Prompt caching: platform.claude.com
Claude Platform docs, API credits for Max and Team plans: platform.claude.com
Claude Help Center, Monthly API credits for Max and Team plans: support.claude.com
Anthropic, Supplemental Credit Terms: anthropic.com/legal/credit-terms
OpenAI, API pricing (retrieved Oct 7 2026): developers.openai.com
VentureBeat, Anthropic launches Claude Haiku 5.5 with 90% API price reduction, matching GPT-6 Luna (Oct 7 2026): venturebeat.com
Unite.AI, Anthropic slashes Claude Sonnet 5.5 cache-read cost by 50% (Oct 7 2026): unite.ai

New Order Form decoders land in the weekly issue first. Subscribe.

Take it to the room

The short list from this piece

Pulled from the argument above, written so you can read it out in a pipeline or board review.

Frequently asked questions

How much does Claude Haiku 5.5 cost?: $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Over 100,000 tokens, $0.50 input and $2.50 output. Cache reads are $0.01 under the line and $0.05 over. Batch is 50% off.
How does Haiku 5.5 compare with GPT-6 Luna?: Same short-prompt rates: $0.10 input and $0.50 output per million tokens. Luna's short context runs to 272,000 input tokens and its long-context rates are $0.20 and $0.75. Haiku 5.5 switches to 5x rates over 100,000 tokens. The two models count tokens differently, so test on your own text.
What does a million lead decisions cost on Haiku 5.5?: At 1,000 input tokens and 50 output tokens per decision, about $125 at list, or $0.125 per 1,000 decisions. Haiku 4.5 costs $1,250 for the same token counts. Batch cuts Haiku 5.5 to $62.50.
What happens over 100,000 tokens?: Every Haiku 5.5 rate rises 5x. A brief with 101,000 input tokens and 2,000 output tokens costs $0.0555, against $0.0109 at 99,000 input tokens. Anthropic has not published whether cached tokens count toward the line.
Do Claude Team plans include API credits?: Yes, from the week of Oct 7 2026. $20 per Standard seat and $100 per Premium seat each month, pooled and capped at $500. Max 5x gets $100 and Max 20x gets $200. Free, Pro, and Enterprise are not eligible. New subscribers wait 7 days.
Do unused Claude API credits roll over?: No. They expire at the end of each billing cycle, or monthly on annual plans. If the organization has no purchased credits, API requests stop when the credits run out.
What should go on the Order Form?: A price lock on the under-100,000 rates, a written threshold definition, whole-request or overage billing, a prompt size cap, a tokenizer test, credits treated as pilot money, credit scope by cloud, workspace spend caps, and the model retirement date.

Share this page

Posting to Instagram or TikTok? Copy the link, it carries the title, summary and share image.

Get the next Reality Check before you sign the order form.

Keep reading

Reality Check

AI Review Time Is a Line Item. Price It and Name the Owner.

On TextQL's Argo-Bench the best AI model fully solved 34.8% of 210 data tasks. Put review time and a named owner into your AI business case.

Reality Check

Call Center Headcount Is Falling. Can You Prove AI Did It?

Revelio Labs says global call-center headcount is 5.3% below its Dec 2023 peak. Revelio: "The decline is surely multifaceted." How to test your own org.