Haiku 5.5 vs GPT-6 Luna: API pricing by prompt

According to APITariff, Haiku 5.5 vs GPT-6 Luna API pricing is identical up to 100,000 prompt tokens: both list $0.10 per 1M input tokens and $0.50 per 1M output tokens on Anthropic’s and OpenAI’s pricing pages, read 2026-10-10. Above 100,000 tokens Haiku 5.5 bills the request at $0.50 and $2.50 per 1M tokens, 5 times its base rate, while GPT-6 Luna holds $0.10 and $0.50 until 272,000 tokens. Computed from those rates, a call with a 150,000-token prompt and 1,000 output tokens bills $0.0775 on Haiku 5.5 and $0.0155 on GPT-6 Luna. Haiku 5.5 and GPT-6 Luna each come to $250 over 1,000,000 calls of 1,000 input and 300 output tokens.

Up to 100,000 tokens the two tie. From 100,001 to 272,000 tokens GPT-6 Luna bills one fifth of Haiku 5.5, and above 272,000 it depends on whether OpenAI reprices the whole request. We compared 12 price fields and counted 0 where Haiku 5.5 is lower: 6 tie and 6 favour Luna.

The OpenRouter and kie.ai rows we fetch hold one rate per model, so a database built on them shows Haiku 5.5 at $0.10 and $0.50 and misses the 100,000-token tier. This page prices tokens and scores no model on quality.

Last updated 2026-10-10 · 2 models compared on 12 price fields · sources: Anthropic and OpenAI pricing pages, APITariff fetch of OpenRouter and kie.ai, 2026-10-10.

Haiku 5.5 vs GPT-6 Luna API pricing, side by side

The two models match on all 6 price fields at their base tier and differ on all 6 above it. The columns pair each model’s own tiers, so the higher tiers start at different prompt sizes: 100,000 tokens for Haiku 5.5, 272,000 for GPT-6 Luna.

$ per 1M tokens	Haiku 5.5, up to 100K	Haiku 5.5, over 100K	GPT-6 Luna, up to 272K	GPT-6 Luna, over 272K
Input	0.10	0.50	0.10	0.20
Output	0.50	2.50	0.50	0.75
Cache read	0.01	0.05	0.01	0.02
Cache write	0.125	0.625	0.125	0.25
Batch input	0.05	0.25	0.05	0.10
Batch output	0.25	1.25	0.25	0.375

Sources: Anthropic pricing and OpenAI pricing, standard and batch tables, both read 2026-10-10. The Haiku 5.5 cache write is the 5-minute rate; the 1-hour rate is $0.20 up to 100K and $1.00 above.

OpenAI lists a single cache-writes column, which the table sets beside Haiku’s 5-minute rate. OpenRouter lists a 1,000,000-token window for Haiku 5.5 and 1,050,000 for GPT-6 Luna, fetched 2026-10-10. OpenRouter also lists openai/gpt-6-luna-pro at $0.10 and $0.50, but OpenAI’s pricing page has no such row, so it is not compared.

How much do Haiku 5.5 and GPT-6 Luna cost at 10k, 100k and 1M calls?

At 1,000 input and 300 output tokens per call both models bill $250 for 1,000,000 calls, and $125 through batch. At a 150,000-token prompt with 1,000 output tokens Haiku 5.5 bills $77,500 for 1,000,000 calls and GPT-6 Luna $15,500. To price your own prompt sizes, enter them in the token calculator.

Workload per call	Model	10k calls	100k calls	1M calls
1,000 in / 300 out	Haiku 5.5	$2.50	$25	$250
1,000 in / 300 out	GPT-6 Luna	$2.50	$25	$250
150,000 in / 1,000 out	Haiku 5.5	$775	$7,750	$77,500
150,000 in / 1,000 out	GPT-6 Luna	$155	$1,550	$15,500

Standard rates, no caching; each call is priced at the tier its prompt length sets.

What does a 100,001-token prompt cost on Haiku 5.5?

A 100,001-token prompt with 1,000 output tokens bills $0.0525 on Haiku 5.5, 5 times the $0.0105 a 100,000-token prompt costs. A call with 100,000 cached tokens, 10,000 new ones and 1,000 output tokens bills $0.0125 on Haiku 5.5 against $0.0025 on GPT-6 Luna.

Anthropic’s pricing page says “a request whose prompt is over 100,000 tokens pays higher prices”, that “Each request is priced on its own”, and that cache reads and writes count toward prompt length. Anthropic’s launch page, dated 2026-10-07, says 90% of Haiku 4.5 requests had prompts up to 100,000 tokens; your share is unknown, and above the line the same call bills 5 times as much.

Prompt tokens	Haiku 5.5, 10,000 calls	GPT-6 Luna, 10,000 calls	Haiku 5.5 / Luna
50,000	$55	$55	1.00
100,000	$105	$105	1.00
100,001	$525.01	$105	5.00
150,000	$775	$155	5.00
272,000	$1,385	$277	5.00
272,001	$1,385.01	$551.50	2.51

Workload: 1,000 output tokens per call, standard rates, no caching, tier set by the prompt length of each request. OpenAI’s page defines its line as “>272K input tokens” and does not say in words whether the whole request moves to the higher rate; the 272,001 row assumes it does, as for Haiku 5.5.

Where GPT-6 Luna wins on price

GPT-6 Luna is lower than Haiku 5.5 on 6 of 12 price fields, all 6 in the higher-tier columns (over 100K for Haiku 5.5, over 272K for Luna), by 60% to 70%. In those columns input is $0.20 against $0.50, output $0.75 against $2.50, and batch output $0.375 against $1.25.

Between 100,001 and 272,000 prompt tokens Luna is still at its base rates, so Haiku 5.5 costs 5 times as much; at 272,001, if OpenAI bills the whole request as assumed above, the ratio falls to 2.51. A batch request with a 150,000-token prompt and 1,000 output tokens bills $0.00775 on Luna and $0.03875 on Haiku 5.5. Luna’s OpenRouter row held $0.10 and $0.50 across 17 fetches since 2026-09-23, and it lists the larger window, 1,050,000 tokens against 1,000,000.

Where Haiku 5.5 wins on price

On none of the 12 fields: 6 tie and 6 favour GPT-6 Luna. Any reason to pick Haiku 5.5 sits outside price data. We have not run either model, so this page makes no claim on quality, speed, accuracy or where each is sold; Anthropic’s statements on those are Anthropic’s, not APITariff measurements.

How much cheaper is Haiku 5.5 than Haiku 4.5?

Haiku 5.5 bills 90% less than Haiku 4.5 up to 100,000 prompt tokens and 50% less above, computed from Anthropic’s rates ($1 input and $5 output on Haiku 4.5), the same two figures Anthropic prints on its launch page.

A call of 1,000 input and 300 output tokens bills $0.0025 on Haiku 4.5 and $0.00025 on Haiku 5.5; at a 150,000-token prompt with 1,000 output tokens it bills $0.155 against $0.0775.

Per OpenRouter’s listing Haiku 4.5 has a 200,000-token window. A move from Haiku 4.5 can cross the 100,000-token line without any change to the text, because Haiku 5.5’s newer tokenizer counts the same text as more tokens.

Does the 100,000-token line move with the tokenizer?

Probably. Anthropic’s pricing page puts the newer tokenizer of Claude 4.7 and later at “approximately 30% more tokens for the same text”, and its launch page says Haiku 5.5 “uses slightly more tokens per task”. If the 30% holds, 80,000 Haiku 4.5 tokens become about 104,000 on Haiku 5.5, over the line.

On our estimate the line sits at about 76,900 Haiku 4.5 tokens. That is derived from Anthropic’s approximate figure, not measured, and the exact count depends on the text. GPT-6 Luna uses OpenAI’s tokenizer and we have not counted one text on both, so count your own prompt with each provider’s token counter before you price it.

Is Claude Haiku 5.5 free?

Not as a standing free tier on the pages we read. Anthropic’s pricing FAQ (read 2026-10-10) says new users get “a small amount of free credits to test the API” and gives no amount. Anthropic’s launch page, dated 2026-10-07, adds a monthly API credit for Claude Max and Team subscribers, also without an amount.

The pricing page lists no open-weights release for Haiku 5.5. We did not check other sources, so open-weights status is not covered here.

The rule: which costs less at your prompt size?

Up to 100,000 prompt tokens, Haiku 5.5 and GPT-6 Luna bill the same, so price does not decide. From 100,001 to 272,000 tokens GPT-6 Luna bills one fifth of Haiku 5.5. Above 272,000 tokens its bill falls to about 40% of Haiku 5.5’s at 1,000 output tokens per call, if OpenAI reprices the whole request.

Batch follows the same lines: a tie to 100,000 tokens, then Luna at one fifth to 272,000. Anthropic prints the 100,000-token line and the request-level rule; OpenAI prints the 272,000-token line, and the whole-request reading above it is our assumption. This is price only, on direct rates; whether either model is worth its bill is outside this page, so count your prompt against the two lines before you choose.

Is Haiku 5.5 or GPT-6 Luna cheaper through kie.ai?

GPT-6 Luna: kie.ai lists it at $0.03 input and $0.15 output per 1M tokens, against $0.06 and $0.30 for Haiku 5.5, on pages read 2026-10-10. Those listed rates sit 70% under direct for Luna and 40% under for Haiku 5.5 at a 1,000-token prompt; neither page publishes a long-prompt price, and the Haiku row rests on 1 fetch.

At 1,000,000 calls of 1,000 input and 300 output tokens, Luna through kie.ai bills $75 and Haiku 5.5 bills $150, against $250 each direct.

Model	kie.ai $ in / out per 1M	Direct $ in / out per 1M	Bill via kie.ai, 1M calls	Bill direct, 1M calls	Bill direct batch, 1M calls
Claude Haiku 5.5	0.06 / 0.30	0.10 / 0.50	$150	$250	$125
GPT-6 Luna	0.03 / 0.15	0.10 / 0.50	$75	$250	$125

Sources: kie.ai pages for Haiku 5.5 and GPT-6 Luna, read 2026-10-10; our fetch of the same rows, 2026-10-10. The kie.ai Haiku 5.5 row first appeared in our fetch on 2026-10-10, and kie.ai lists Luna on a combined GPT-6 Sol and Luna page.

Where kie.ai is not cheaper: Haiku 5.5 through Anthropic’s batch API bills $125 for the same 1,000,000 calls, below kie.ai’s $150 listed rate and about $135 at kie.ai’s stated approximately 10% lower effective price. For Luna, kie.ai’s $75 still beats batch at $125.

What is unknown: neither kie.ai page lists a price for prompts over 100,000 tokens (Haiku 5.5) or over 272,000 (GPT-6 Luna), so kie.ai is not a priced route for a 150,000-token prompt, and we have not sent a request through kie.ai for either model. The table uses listed rates; kie.ai’s Luna page ties its roughly 10% lower effective price to high-tier top-ups, and its Haiku page gives no condition. kie.ai is not Anthropic or OpenAI, and we have not read its data-retention terms.

Affiliate link: if you sign up through it we may earn a commission. The comparison above uses direct rates only; the rates in this section come from kie.ai’s own pages and our fetch of them, and the link changed none of them. kie.ai pricing shows the full rate list; our kie.ai price page and the kie.ai vs list prices comparison for 31 models go further.

How we compared

The rule used here: take each provider’s own pricing page, pair the two models at their base price, price each call at the tier its prompt length sets, and report 12 fields. Price only; no quality score and no “best”.

OpenRouter’s Haiku 5.5 row carries one rate, $0.10 and $0.50 (run 135, fetched 2026-10-10 08:30 UTC, first read 2026-10-08, unchanged in 2 fetches). The $0.50 and $2.50 rate for longer prompts is on Anthropic’s page only, so a price database that stores one rate per model understates the bill for prompts over 100,000 tokens. Collection details are on the methodology page.

For the earlier pairing, Haiku 4.5 against GPT-5.6 Luna as checked 2026-09-15, see the Claude vs OpenAI API pricing comparison. Claude’s other models sit in the Anthropic API pricing guide, the Claude Opus 5.5 post covers Anthropic’s earlier Opus price cut, and the cheapest LLM API ranking puts these rows beside every text model we fetch.

Cite as: According to APITariff (apitariff.com/compare/haiku-5-5-vs-gpt-6-luna-api-pricing/, read 2026-10-10).

Drafted with AI from APITariff’s own price data, checked and edited by a person. A script draws the chart from the figures in the tables above; no image is generated.

Change log

• 2026-10-10First edition. Standard and batch rates read on the Anthropic and OpenAI pricing pages 2026-10-10; Haiku 5.5 first seen in our fetch 2026-10-08.