Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6 Luna launched about a week apart at identical rates: $0.10 input, $0.50 output, $0.01 cache hits and $0.125 cache writes per 1M tokens. The real price difference only shows up on long prompts.
For prompts under 100K tokens they cost exactly the same. Between 100K and 272K tokens GPT-6 Luna is much cheaper, because Claude Haiku 5.5 jumps to $0.50/$2.50 per 1M while Luna stays at its base rate. Above 272K Luna rises to $0.20/$0.75 — still well below Haiku's long-prompt price.
| Claude Haiku 5.5 | GPT-6 Luna | |
|---|---|---|
| Provider | Anthropic | OpenAI |
| Tier | Budget | Budget |
| Input / 1M tokens | $0.10 | $0.10 |
| Output / 1M tokens | $0.50 | $0.50 |
| Cache hit / 1M | $0.01 | $0.01 |
| Cache write / 1M | $0.125 | $0.125 |
| Context window | 1M | 1M |
| Notes | Released Oct 7, 2026; 90% cheaper than Haiku 4.5 with a 1M context window (for prompts ≤100K tokens); >100K tokens: $0.50/$2.50, cache read $0.05, cache write $0.625 | New (Sep 29, 2026 DevDay), fastest & cheapest GPT-6-family tier; >272K tokens: $0.20/$0.75, cache $0.02 |
Both charge $0.10 per 1M input tokens and $0.50 per 1M output tokens.
| Workload | Claude Haiku 5.5 | GPT-6 Luna | Difference |
|---|---|---|---|
| Support chatbot 10,000 conversations × (2,000 input + 400 output tokens) | $4.00 | $4.00 | — |
| Document Q&A 1,000 questions × (30,000 input + 800 output tokens) | $3.40 | $3.40 | — |
| Coding agent session 4M input tokens (90% cache hits) + 150,000 output tokens | $0.15 | $0.15 | — |
Standard rates per 1M tokens from data on this site. The coding-agent row assumes 90% of input is read from the prompt cache (full input price where a model has no cache rate). Cache-write fees, batch discounts and long-context surcharges listed in the notes are not included. Updated November 2, 2026.
your prompts stay under 100K tokens, you already build on Anthropic's API or tooling, or you prefer Claude-family behaviour on your evals — at that size you pay nothing extra for it.
you regularly send prompts over 100K tokens — large documents, long chat histories, repository context — where Luna's later and smaller long-context step saves real money.
Anthropic prices Haiku 5.5 by prompt length: a request whose prompt is over 100,000 tokens pays $0.50 input, $2.50 output, $0.05 cache read and $0.625 cache write per 1M — and its pricing page says cache reads and writes count toward that length, so a mostly cached 120K prompt still pays the higher rate. OpenAI's GPT-6 Luna steps up only above 272K input tokens, to $0.20/$0.75 with $0.02 cache hits. On a 150K-token prompt with a 1K answer, Haiku's bill is five times Luna's.
Both charge $0.125 per 1M to write a cache at the base tier, but the lifetimes differ: Anthropic's standard write is a 5-minute cache, while OpenAI's matching price buys a 30-minute cache. Longer-lived caches cost more on both: Anthropic's 1-hour write is 2× base input, and OpenAI's 12-hour write is double its 30-minute price. For bursty traffic with gaps of several minutes, Luna's longer default window means fewer re-writes.
Both providers halve all rates for their Batch APIs, including Haiku's over-100K tier, so the relative picture holds for offline work. With base prices identical, choose on quality for your task, latency and which SDK your stack already uses — and only let price decide if long prompts are common.
At base rates, yes: $0.10 input, $0.50 output and $0.01 cache hits per 1M on both. They differ only for long prompts.
When a request's prompt exceeds 100K tokens, counting cached tokens. The whole request is then billed at $0.50 input and $2.50 output per 1M.
Above 272K input tokens, where it bills $0.20 input, $0.75 output and $0.02 cache hits per 1M.
Yes, both list 1M tokens.
How we choose: prices come from each vendor's official pricing page and are re-checked every week. We take no affiliate fees, so rankings aren't for sale.