Independent comparison · prices verified Oct 9, 2026
Haiku 5.5
vs Luna 6
Same price per token — until your prompt passes 100K tokens. Then Haiku costs five times as much.
USD per million tokens, input / output. The whole request moves to the higher rate.
Short, demanding prompts under 100K tokens.
Same rate as Luna, ahead on every benchmark in Anthropic’s launch comparison.
Long prompts and routine high-volume work.
Keeps its base price to 272K tokens, where Haiku has already jumped 5×.
What will it cost you?
Customize tokens and volume
Everything you send — system prompt, history, documents, tools. Cached tokens count toward the threshold too.
$0.10 / $0.50
Identical input / output price per million tokens on both, cache reads $0.01.
43 vs 38
Artificial Analysis Intelligence Index at max effort. FrontierCode: 46.4 vs 42.4. Source
~3×
The output tokens Haiku used to run that index (435M vs 144M) — and output is the expensive side.
Which one, by workload
Short, hard tasks under 100K
Haiku 5.5Same price per token, ahead on every benchmark that has a Luna score.
Long documents, 100K–272K
Luna 6Haiku reprices the whole request; Luna stays at base and is about 5× cheaper.
Huge prompts over 272K
Luna 6Both reprice, but Luna’s input is $0.20 vs Haiku’s $0.50 per million.
High-volume classification
EitherIdentical bill at a few thousand tokens. Pick on accuracy from a quick eval.
Verbose reasoning at max effort
Luna 6Haiku used ~3× the output tokens on the AA index, so its bill can be higher even under 100K.
Overnight batch jobs
EitherBoth are half price in batch; the same thresholds decide the winner.
Specs and prices
USD per million tokens. Haiku pricing · Luna pricing
Both bill the whole request at the higher tier once the prompt crosses the threshold; Anthropic counts cache reads and writes toward its 100K. Sources: Anthropic pricing, Claude Haiku 5.5 model overview, OpenAI GPT-6 Luna model page, OpenAI API pricing.
Questions
Is GPT-6 Luna cheaper than Claude Haiku 5.5?
Only for long prompts. Both list at $0.10 per million input tokens and $0.50 per million output tokens, with cache reads at $0.01, as long as the prompt stays at or under 100K tokens. Above 100K, Anthropic bills the whole Haiku 5.5 request at $0.50 input and $2.50 output, while Luna keeps its base price up to 272K tokens and then moves to $0.20 input and $0.75 output. A 150K-token prompt with 2K output costs about $0.080 on Haiku and $0.016 on Luna.
Which is better, Haiku 5.5 or Luna 6?
Haiku 5.5, on the published numbers. In Anthropic’s launch comparison it leads every benchmark that has a Luna score (FrontierCode 1.1: 46.4% vs 42.4%), and the independent Artificial Analysis Intelligence Index agrees at max effort, 43 vs 38. Haiku used about three times as many output tokens to run that index, though (435M vs 144M). Short, demanding tasks favour Haiku; long prompts and routine high-volume work favour Luna.
Why can Haiku 5.5 cost 12× more than Luna 6?
Two effects multiply. Above 100K prompt tokens Haiku bills the whole request at 5× its base rate, while Luna keeps its base price to 272K. And Haiku tends to write more: it used about 3× Luna’s output tokens to run the Artificial Analysis Intelligence Index. A request just over 100K tokens with 50K output costs about 12× more on Haiku. Under 100K the gap can only come from output length, so it stays near 3× at most. Reports of a 12× gap usually come from long-context agents and single runs; measure your own workload before generalising.
What happens above 100K tokens on Haiku 5.5?
The entire request is billed at the higher tier ($0.50 in, $2.50 out, $0.05 cache read), not just the tokens past the line. The prompt length that decides the tier counts every input token, including cache reads and cache writes. Each request is priced on its own.
Does GPT-6 Luna support prompt caching?
Yes. OpenAI lists cached input for Luna at $0.01 per million tokens and cache writes at $0.125 per million on the standard tier (prompts up to 272K), the same as Haiku 5.5 under 100K. Some comparison pages list Luna caching as unpublished; that is out of date.
Which model is better for long documents?
Luna, on cost. Both accept roughly a million tokens of context (1M for Haiku, 1.05M for Luna), but between 100K and 272K tokens Luna is about five times cheaper per request, and above 272K it is still about 2.5 times cheaper on input.
Do both models offer batch pricing?
Yes. Anthropic’s Message Batches API and OpenAI’s Batch and Flex tiers are each half the standard price, and the long-prompt thresholds still apply. Turn on the Batch toggle in the calculator to see the effect.
Is Haiku 5.5 cheaper than Haiku 4.5?
Much cheaper on list price: Haiku 4.5 is $1 input and $5 output per million tokens, Haiku 5.5 is $0.10 and $0.50 for prompts up to 100K tokens. Above 100K, Haiku 5.5 is $0.50 and $2.50, still half of Haiku 4.5.
Can I keep data in the US?
On the Claude API you can pin Haiku 5.5 to US-only inference with the inference_geo parameter, which multiplies every price by 1.1. OpenAI lists a 10% premium for regional processing on Luna where it is available; check which regions your account can use.