GPT-6 Luna vs Claude Haiku 5.5: price, speed, context, and where your prompts go

The Morse teamPublished October 30, 2026Facts checked October 23, 20267 min read

GPT-6 Luna and Claude Haiku 5.5 cost the same per token for short prompts, $0.10 per million input tokens and $0.50 per million output tokens, but Claude Haiku 5.5 charges five times that once a prompt passes 100,000 tokens, while GPT-6 Luna holds its base price up to 272,000. Choose GPT-6 Luna if you send long documents through the API, choose Claude Haiku 5.5 if Anthropic's published benchmarks match your work, and choose Morse if you want to chat with both in one account, and use GPT and Claude in a mode where neither company knows it's you.

We read OpenAI's GPT-6 Luna model page, its API changelog, its October 2026 safety update, and its API data controls, and Anthropic's Haiku 5.5 announcement, model page, pricing, system card, and retention pages, on October 23, 2026. OpenAI's announcement post and its consumer help center block automated reads, so nothing here relies on them. Every benchmark score below is Anthropic's; OpenAI's developer pages we read don't publish any.

GPT-6 Luna vs Claude Haiku 5.5 at a glance

As of October 23, 2026. Every OpenAI and Anthropic fact links to its maker's own page.

Which is cheaper, GPT-6 Luna or Claude Haiku 5.5?

It depends on prompt length. Below 100,000 tokens the price lists match: OpenAI charges $0.10 per million input tokens, $0.01 cached, and $0.50 output for GPT-6 Luna (OpenAI), and Anthropic charges $0.10, $0.01 for cache reads, and $0.50 for Claude Haiku 5.5 (Anthropic). Both take 50% off for batch jobs.

Long prompts split them. Anthropic prices a Haiku 5.5 request whose prompt is over 100,000 tokens at $0.50 in and $2.50 out, and counts cache reads and writes toward that threshold (Anthropic). Haiku 5.5 is the only current Claude model priced this way; Anthropic's pricing page says the others bill a 900,000-token request at the same rate as a 9,000-token one. GPT-6 Luna has a surcharge too, at a higher threshold. Above 272,000 input tokens, OpenAI bills the whole request at twice the input rate and 1.5 times the output rate (OpenAI).

We worked out what one request costs at three prompt sizes, each with a 2,000-token reply, from those two price lists, with no caching or batch discount.

Token counts aren't the same across the two companies, so the same document won't come to the same number in each. Anthropic says Haiku 5.5's tokenizer counts the same text as about 30 percent more tokens than Haiku 4.5's did (Anthropic). Neither company compares its count with the other's, so run your own documents through both before you budget.

Two smaller charges sit on top. OpenAI adds 10% for regional processing where it's offered and charges double for Fast mode (OpenAI). Anthropic charges 1.1 times for US-only inference, including on Haiku 5.5's long-prompt prices (Anthropic).

Which is faster?

Neither company publishes a tokens-per-second figure for its model on the pages we read. Anthropic rates Haiku 5.5's latency "Fastest" in its current lineup (Anthropic) and calls it "our fastest model to date" at standard speed (Anthropic). Customers quoted in that announcement report lower latency than Haiku 4.5, not against GPT-6 Luna.

OpenAI's model page gives GPT-6 Luna no speed rating. It sells speed as an option instead: Fast mode at twice the price, and the Decisions API, which OpenAI's changelog says turns text and images into typed answers "10x faster than the Responses API" and made generally available with gpt-6-luna on October 22, 2026 (OpenAI).

Is Claude Haiku 5.5 better than GPT-6 Luna?

On Anthropic's own scoreboard, yes, on every test where it lists both. These are the GPT-6 Luna numbers in Anthropic's Haiku 5.5 announcement. OpenAI's developer pages we read publish none to set against them, and we haven't run the tests ourselves.

• OSWorld 2.1, offline subset (using a computer): Haiku 5.5 72.4%, GPT-6 Luna 48.9%.
• Terminal-Bench 4.0 (command-line tasks): 39.2% against 16.4%.
• FrontierCode 1.1, Main: 46.4% against 42.4%.
• GDPval-AA v2.1 (work tasks, Elo): 1620 against 1437.
• Chartography, no tools: 46.4% against 29.1%.

Anthropic's own Sonnet 5.5 beats Haiku 5.5 on every one of those rows, at 20 times the input price (Anthropic).

How do they handle safety and prompt injection?

Both makers publish safety results. OpenAI's October 2026 update treats GPT-6 Luna as High capability in cybersecurity and in biology and chemistry under its Preparedness Framework, and reports 99.79% robustness on its instruction hierarchy tests. Anthropic's system card reports that attackers given 15 tries at Gray Swan's indirect prompt injection benchmark succeeded 7.1% of the time against Haiku 5.5, down from 83.2% for Haiku 4.5, with most of what remains in computer use.

Anthropic also says Haiku 5.5's cyber classifiers are "significantly narrower" than those on its bigger models, and that its safety classifiers can now decline a request, which API code sees as a refusal stop reason (Anthropic).

What happens to your prompts on the API and in the apps?

On the APIs, neither company trains on what you send by default. OpenAI says "data sent to the OpenAI API is not used to train or improve OpenAI models" unless you opt in, keeps abuse monitoring logs "for up to 30 days, unless longer retention is required by law," and keeps Responses API state for 30 days by default (OpenAI). Zero data retention needs OpenAI's "prior approval."

Anthropic's two pages on this read differently. Its API docs say "conversation content (your prompts and Claude's outputs) is not retained by default" for models like Haiku 5.5 (Anthropic). Its privacy center says Anthropic deletes API inputs and outputs "within 30 days of receipt or generation," and keeps flagged ones for up to 2 years (Anthropic).

The consumer apps follow different rules. OpenAI's safety update says GPT-6 is launching "to everyone in ChatGPT, across free and paid plans," where GPT-6 Luna replaces GPT-5.6 Luna (OpenAI), and ChatGPT's own data settings apply there; we cover them in does ChatGPT train on your data. Claude's Free plan includes "Haiku," without naming the version (Anthropic). If you allow training in the Claude app, Anthropic may keep your chats in de-identified form "for up to 5 years" (Anthropic), and Haiku 5.5's system card lists data "which users have explicitly permitted for training" among its sources. Our guide to whether Claude is private has the settings.

How Morse lets you use both without an account at either company

In ChatGPT or the Claude app, every prompt sits under your own account, tied to your email and payment details, under that company's retention rules. Morse has three privacy modes, chosen per chat. Proprietary models like GPT and Claude run in Anonymous mode. Our proxy removes your account, email, and IP address and sends the message under Morse's own account, so OpenAI or Anthropic reads the prompt without knowing it's you. The provider may keep it for a limited time to check for abuse, under its own policy, so leave out names you'd rather not send. Our guide to using ChatGPT anonymously explains the method.

Your chat history stays in your browser, encrypted, and Morse's servers don't keep messages or log what you write. Morse doesn't train AI on your content, sell your data, or show ads. Clearing your browser deletes your chats, and Morse can't recover them, so keep a copy of anything you need.

For questions you'd rather neither lab saw, switch to an open-weight model. Private mode runs models like GLM, Kimi, and DeepSeek, and Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Our page on GLM 5.3 Flash covers one of those models. On Pro, E2EE chats encrypt your message in the browser so only a verified enclave can read it, and the guarantee rests on the enclave hardware and attestation working as designed.

Which should you choose?

• Choose GPT-6 Luna if you send prompts between 100,000 and 272,000 tokens, want reasoning switched fully off with effort set to none, or want OpenAI's built-in API tools: web search, file search, code interpreter, and the Decisions API.
• Choose Claude Haiku 5.5 if Anthropic's benchmark rows look like your work, especially computer use and command-line tasks, you want a June 2026 knowledge cutoff, or you need it through Amazon Bedrock, Google Cloud, or Microsoft Foundry, where Anthropic lists it.
• Choose Morse if you want to chat with GPT-6 Luna and Claude Haiku 5.5 without an OpenAI or Anthropic account, keep the history only in your browser, and move to E2EE when a question calls for it.

Send a message from the Morse home page without an account, or start free for 15 messages a day with both models.

Morse publishes this comparison. We've tried to keep it fair; if something's wrong or out of date, email hello@trymorse.ai.

GPT is a trademark of OpenAI. Morse is not affiliated with or endorsed by OpenAI.

Claude is a trademark of Anthropic. Morse is not affiliated with or endorsed by Anthropic.