Claude Haiku 5.5 vs GPT-6 Luna (Luna 6): Benchmarks, Pricing & Context Window haiku55vsluna6
haiku55vsluna6 is a side-by-side breakdown of OpenAI's GPT-6 Luna and Anthropic's Claude Haiku 5.5 across context window, output ceiling, API pricing, cache rates, published benchmarks, declared capabilities and release history. Spelled out in full, the matchup is Claude Haiku 5.5 vs GPT-6 Luna (Haiku 5.5 vs Luna 6) — the same head-to-head under both namings. Every figure below reflects what the two vendors publish; anything they leave undisclosed is shown as not published rather than estimated.
haiku55vsluna6 at a glance
- Topichaiku55vsluna6 — Claude Haiku 5.5 (Anthropic) and GPT-6 Luna (OpenAI), compared attribute by attribute.
- IdenticalInput $0.10 and output $0.50 per million tokens, a 128K output ceiling, and closed-source licensing on both sides.
- Luna leads1.05M context window versus 1M; the only model with an intelligence index score (37); declares structured output and vision.
- Haiku leadsFrontierCode 46.4 vs 42.4 — the only like-for-like benchmark; publishes cache rates; ships later with a later knowledge cut-off.
Head-to-head specifications
Context window, drawn to scale
Published benchmark results
Percentage-scale benchmarks
Elo and index metrics
Every benchmark row
| Benchmark | GPT-6 Luna | Claude Haiku 5.5 | Scope / reporting note |
|---|
Why these numbers cannot be ranked against each other
- Luna's results are all labelled "max reasoning effort." OpenAI attaches that note to every Luna benchmark, so Luna's figures describe its highest-effort configuration.
- Haiku 5.5's results are all "Anthropic-reported launch result." They are vendor-reported numbers, not independent reproductions.
- OSWorld is not the same version on both sides. Luna's entry is OSWorld 2.0 (offline set, v2026.08.08); Haiku's is OSWorld 2.1 (offline subset). They sit on the same row but they were not measured on the same task set.
Declared capability coverage
Ten capability flags appear across the two models. Three are shared; the remaining seven are claimed by one model only. These are vendor-declared labels, not measured results.
Supported modalities
| Modality | GPT-6 Luna | Claude Haiku 5.5 | Result |
|---|---|---|---|
| Text | Supported | Supported | Identical |
| Image | Supported | Supported | Identical |
| Voice | Not supported | Not supported | Identical |
| Video | Not supported | Not supported | Identical |
List price and cost modelling
Both models carry an identical list price. The only difference in published pricing is caching: Haiku 5.5 lists cache rates, Luna lists none.
List price per million tokens
| Rate | Luna | Haiku |
|---|---|---|
| Input | $0.10 | $0.10 |
| Output | $0.50 | $0.50 |
| Cache read | not published | $0.01 |
| Cache write | not published | $0.125 |
Cost calculator
input × in-rate + cached × cache-read + output × out-rate.
One-time cache-write costs are not modelled, so a real-world saving would be slightly smaller. Because Luna
publishes no cache rate, cached Luna input is modelled at the full input rate — so any cache hit rate above
zero makes Haiku the cheaper option here. This is arithmetic on published rates, not a
quote or a measurement.
Release and knowledge cut-off
GPT-6 Luna shipped first; Claude Haiku 5.5 shipped later and cites the more recent knowledge cut-off.
haiku55vsluna6 — frequently asked questions
Short answers to the questions this comparison is built to settle. Each answer sticks to what the two vendors actually publish.
Is GPT Luna better than Haiku?
There is no single published answer. The only like-for-like benchmark is FrontierCode, reported by both vendors on the same version: Claude Haiku 5.5 scores 46.4 against GPT-6 Luna's 42.4. Every other benchmark row is published by one vendor only — OpenAI labels its Luna figures "max reasoning effort" and Anthropic labels its Haiku results "launch result" — so no overall winner can be claimed from the published numbers.
Is Luna cheaper than Haiku?
Not at list price: both models publish identical rates of $0.10 per million input tokens and $0.50 per million output tokens. The only published difference is caching — Claude Haiku 5.5 lists a cache read rate of $0.01 and a cache write rate of $0.125, while OpenAI lists no cache rate for GPT-6 Luna. Since cached Haiku input prices at $0.01 against Luna's full $0.10 input rate, any cache hit rate above zero makes Haiku the cheaper option on this arithmetic; with no caching the two cost the same.
Is Haiku deprecated?
No — nothing Anthropic publishes indicates that. Claude Haiku 5.5 carries active list pricing ($0.10 / $0.50 per million tokens plus cache rates), eight declared capability flags, an October 7, 2026 release date and a June 30, 2026 knowledge cut-off — the more recent of the two models compared here. Neither vendor publishes a deprecation or retirement date for either model.
Which is better, Opus or Haiku?
That question cannot be settled from the published numbers here: this comparison covers Claude Haiku 5.5 vs GPT-6 Luna and contains no Claude Opus figures. What Anthropic publishes for Haiku 5.5 in this comparison is a positioning claim — its fastest and most cost-efficient current model — plus the pricing, benchmarks and capability flags listed above. Anthropic's published Opus results would need to be compared separately.