- List price: Luna $0.20 / $1.20 · Haiku 4.5 $1 / $5 per 1M (verify vendors)
- Measured 2026-07-31: same 40-SKU monitor task — both correct; Luna $0.000141 vs Haiku $0.000887 (~6.3×)
- Pool proof: both are tier-1 / Free-accessible in CloudAxis·CloudyBot; Auto Mode itself is fixed to Luna
- Pick Haiku when you need Anthropic-shaped behavior, Claude tooling prefs, or your evals beat Luna on the task
- CloudAxis credits: Luna ≈ 2.5× · Haiku ≈ 10× — Haiku is the dearer Free-tier option on our meters too
This is the Western cheap-tier pairing to our Luna vs DeepSeek post. Full Luna card: complete guide. Cost-per-task method: measured Luna/Terra/Sol.
Price and positioning
Haiku 4.5 is Anthropic’s speed/cost lane. Luna is OpenAI’s after the July 30 cut. On list rates Luna undercuts Haiku on both input and output — roughly 5× cheaper input and ~4× cheaper output before you count token-shape differences.
- Luna
gpt-5.6-luna— $0.20 / $1.20 · cached $0.02 · ~1.05M context · text+image in - Haiku 4.5
claude-haiku-4-5-20251001— $1 / $5 · Anthropic Messages API · strong for Claude-ecosystem apps
Measured completed task
Same prompt family as our cost-per-task series: 40-row SKU snapshot → JSON with cheapest SKU/vendor/price/spread/n. Luna via Chat Completions (reasoning_effort: none); Haiku via Anthropic Messages. Date: 2026-07-31.
- Luna: 501 in / 34 out · $0.000141 · 1074 ms · correct raw JSON
- Haiku 4.5: 562 in / 65 out · $0.000887 · 905 ms · correct JSON in markdown fences
Haiku used more tokens (fenced JSON) and higher rates — about 6.3× the Luna bill for the same answer. Repro notes in sites/cloudaxis/content/gpt-5-6-luna-vs-haiku-measurement.mjs.
This is one task, one run each — not a claim Haiku never wins. It is a claim that on this extraction shape, Luna is the cheaper completed task.
When to choose which
Practical defaults:
- Choose Luna for volume OpenAI-shaped agents, vision-capable cheap tier, Luna-without-Plus product stories, and cost-per-completed-task floors
- Choose Haiku 4.5 when Claude tone/tools/evals win, or procurement wants Anthropic on the invoice
- On CloudAxis: Auto Mode is Luna — no Haiku dropdown. Both remain in the hosted pool for platform routing/fallbacks
Related reading
Frequently asked questions
Is GPT-5.6 Luna cheaper than Claude Haiku 4.5?
On July 2026 list rates, yes — Luna is $0.20/$1.20 vs Haiku’s $1/$5 per 1M. Our SKU-monitor microbench also cost less on Luna for the same correct answer.
Are both models available on CloudAxis Free?
Both are Free-tier in the hosted pool. Auto Mode itself is fixed to GPT-5.6 Luna on every plan — there is no manual Haiku picker.
When should I still use Haiku?
When your evals prefer Claude, you need Anthropic-native features, or vendor constraints require Anthropic. Cost alone is not the only scoreboard.
How do CloudAxis credits compare?
Approximate multipliers: Luna ~2.5×, Haiku 4.5 ~10×. Haiku burns more credits per token on our meters as well as on vendor list rates.
GPT-5.6 Luna complete guide · Luna vs DeepSeek · Cost per completed task · Luna on CloudAxis Free · AI cost calculator