Claude Haiku 5.5 vs GPT-6 Luna is the cleanest head-to-head in AI right now: both cost $0.10 per million input tokens and $0.50 per million output tokens, both handle about a million tokens of context, and both are their lab's fast, cheap model for high-volume work. The short answer: for prompts under 100,000 tokens the price is a tie, Anthropic's own tests put Haiku 5.5 ahead on agentic and computer-use tasks, and GPT-6 Luna is the cheaper choice for long prompts and tends to use fewer tokens per task.
Anthropic released Claude Haiku 5.5 on October 7, 2026. OpenAI released GPT-6 Luna on September 22, and from October 8 it powers GPT-6 in ChatGPT's Chat tab for Free and Go users. This guide sets out what each lab has actually published, as of October 8, 2026.
Claude Haiku 5.5 vs GPT-6 Luna at a glance
| Item | Claude Haiku 5.5 | GPT-6 Luna |
|---|---|---|
| Released | October 7, 2026 | September 22, 2026 |
| API model ID | claude-haiku-5-5 | gpt-6-luna |
| Input / output price (per million tokens) | $0.10 / $0.50 for prompts up to 100K tokens | $0.10 / $0.50 for prompts up to 272K tokens |
| Long-prompt price | $0.50 / $2.50 over 100K tokens | $0.20 / $0.75 over 272K tokens |
| Cache reads / cache writes | $0.01 / $0.125 (up to 100K) | $0.01 / $0.125 |
| Batch discount | 50% | 50% (Batch and Flex) |
| Context window | 1M tokens | 1,050,000 tokens (922,000 max input) |
| Max output | 128K tokens | 128K tokens |
| Knowledge cutoff | June 2026 | May 18, 2026 |
| Effort settings | Yes, default medium | none to max, default medium |
| Input types | Text and images | Text and images |
The Haiku figures come from Anthropic's models overview and pricing docs. The Luna figures come from OpenAI's GPT-6 Luna model page.
Said aloud: both models cost ten cents per million input tokens and fifty cents per million output tokens on short prompts, and both cache reads cost one cent per million. The difference is what happens to long prompts.
Pricing: a tie until 100,000 tokens
Anthropic prices Haiku 5.5 by prompt length. A request whose prompt is over 100,000 tokens pays five times the base rates: fifty cents in and two dollars fifty out per million tokens. Anthropic says prompts of up to 100,000 tokens made up around 90 percent of requests to its previous Haiku model, which is why it built the price around that line.
OpenAI's threshold sits much higher. GPT-6 Luna keeps its base price until a prompt passes 272,000 input tokens. Above that, the whole request is billed at twice the input and cache rates and one and a half times the output rate, so twenty cents in and seventy-five cents out.
Here is what that means for three example requests, worked out from the published list prices:
| Example request | Claude Haiku 5.5 | GPT-6 Luna |
|---|---|---|
| 2,000-token prompt, 200-token answer | $0.0003 | $0.0003 |
| 150,000-token prompt, 2,000-token answer | $0.08 | $0.016 |
| 300,000-token prompt, 5,000-token answer | $0.1625 | $0.064 |
In words: a short classification call costs the same three hundredths of a cent on either model. A 150,000-token document costs about eight cents on Haiku 5.5 and under two cents on Luna, five times less. At 300,000 tokens Luna's own surcharge kicks in and the gap narrows to roughly two and a half times.
Two caveats. First, Anthropic notes that Haiku 5.5 has an updated tokenizer that "uses slightly more tokens per task," so identical text can count as more tokens. Second, the independent testers at Artificial Analysis found that Haiku 5.5 at max effort used about 162,000 output tokens per task on their Intelligence Index, roughly three times GPT-6 Luna at max effort. Same price per token does not mean the same bill per job, so measure on your own workload.
Both labs also cut prices compared with their previous small models. Anthropic says Haiku 5.5 costs around 75 percent less to run than Haiku 4.5 on average. OpenAI cut Luna's price by 50 percent compared with GPT-5.6 Luna's promotional pricing, per its GPT-6 Sol and Luna launch post.
Performance: what each lab claims
Anthropic did something unusual: it put GPT-6 Luna in the comparison table of its own launch post. These are Anthropic's numbers, run by Anthropic, so treat them as vendor claims rather than neutral results.
| Task (Anthropic's launch table) | Haiku 5.5 | GPT-6 Luna |
|---|---|---|
| Computer use, OSWorld 2.1 (offline subset) | 72.4% | 48.9% |
| Agentic coding, Terminal-Bench 4.0 | 39.2% | 16.4% |
| Coding, FrontierCode 1.1 (Main) | 46.4% | 42.4% |
| Knowledge work, GDPval-AA v2.1 (score) | 1620 | 1437 |
In words: Anthropic reports Haiku 5.5 well ahead of Luna on computer use and command-line coding, and narrowly ahead on mergeable code and professional knowledge work.
OpenAI has not published a direct comparison with Haiku 5.5. For Luna, it says GPT-6 Luna at higher effort matches GPT-5.6 Sol on its internal factuality evaluation at about a hundredth of the cost, and that on business-workflow tasks Luna at high effort improves on its predecessor by 5.4 percentage points at 58 percent lower cost per task.
Independent results so far point the same way with a twist. Artificial Analysis scores Haiku 5.5 at 43 on its Intelligence Index at max effort against 38 for Luna at max effort, but notes the heavier token use above. Its numbers are early: it says its cost figures for Haiku 5.5 do not yet include the long-prompt price step.
Speed, effort and where each one fits
Anthropic calls Haiku 5.5 its fastest model to date at standard speed (Opus models in Fast Mode are quicker). It is the first Haiku with an adjustable effort setting, so you can trade cost for intelligence the way you can with Sonnet and Opus. Anthropic pitches it for summaries, compaction, database queries, classification, live customer support, browser use, and as a fast subagent working under Opus 5.5 or Sonnet 5.5. It is explicit that Sonnet 5.5 and Opus 5.5 "remain better choices for complex agentic coding tasks."
OpenAI describes GPT-6 Luna as its "most efficient model for focused, high-volume tasks." Its effort setting runs from none to max, which matters if you want a plain, no-reasoning answer for the cheapest possible call. Luna supports the Batch, Flex and Responses APIs, OpenAI's built-in tools, and EU data residency.
Claude Haiku 5.5: pros
- Strongest small-model results in Anthropic's computer-use and coding tests
- Adjustable effort, newer knowledge cutoff (June 2026)
- Cache reads at one cent per million on short prompts
- Available on the Anthropic API, Amazon Bedrock, Google Cloud and Microsoft Foundry
Claude Haiku 5.5: cons
- Five-times price jump above 100,000-token prompts
- Tends to use more tokens per task, per Artificial Analysis
- Brand new, so less production history
GPT-6 Luna: pros
- Same base price, held to 272,000-token prompts
- Gentler long-prompt surcharge (two times input, one and a half times output)
- A "none" effort option for the cheapest calls
- Now the default GPT-6 model for ChatGPT Free and Go
GPT-6 Luna: cons
- Behind Haiku 5.5 on Anthropic's agentic and computer-use tests
- Older knowledge cutoff (May 18, 2026)
Can you use them for free?
GPT-6 Luna in ChatGPT. OpenAI's GPT-6 and Intelligent UI post says GPT-6 in ChatGPT's Chat tab is powered by GPT-6 Luna for the Free and Go tiers, with the rollout to those tiers starting October 8, 2026. Plus, Pro, Business and Enterprise get GPT-6 Sol instead, and OpenAI says both are tuned for everyday conversation. The same release adds Intelligent UI, which lets answers include charts, buttons, forms and small interactive tools. Free and Go users could already reach Luna in the ChatGPT desktop app.
Haiku in Claude. Claude's pricing page lists Haiku as included on every plan, Free through Max. On the OpenAI API and the Claude Platform, both models are pay-as-you-go.
Anthropic also announced monthly API credits for subscribers alongside Haiku 5.5: $100 a month for Max 5x, $200 for Max 20x, and up to $500 pooled for Team plans, usable on any Claude model. If you already pay for Claude Max, that can cover a lot of Haiku calls.
Safety and safeguards
Anthropic says Haiku 5.5 shows major improvements over Haiku 4.5 on almost all of its alignment evaluations, with far fewer instances of misaligned behavior and a lower willingness to cooperate with misuse. Its cybersecurity safeguards are stricter than Haiku 4.5's but somewhat looser than those on Anthropic's other recent models: they allow a wider range of defensive work than Sonnet 5.5 but still block penetration testing. Biology safeguards match Sonnet 5.5 and Opus 5, and vetted organizations can apply to the Cyber and Life Sciences Verification Programs. Our guide to Claude Fable vs Opus vs Mythos explains how those programs fit Anthropic's lineup.
OpenAI says GPT-6 Sol and Luna build on the alignment work introduced with GPT-6 Astra and improve on their GPT-5.6 counterparts in its alignment evaluations, including lower rates of misleading claims about their coding work. The policies behind both labs' release decisions are compared in Preparedness Framework vs Responsible Scaling Policy.
Which should you choose?
- Short prompts, mostly under 100,000 tokens: price is a tie, so test quality and total tokens on your own tasks. Pick Haiku 5.5 if your work looks like computer use, browser automation or terminal tasks, where Anthropic reports the biggest lead.
- Long documents between 100,000 and 272,000 tokens: GPT-6 Luna is five times cheaper per token.
- Already on one platform: Haiku 5.5 pairs naturally with Opus 5.5 or Sonnet 5.5 as a subagent; Luna sits in the same API as GPT-6 Astra and Sol.
- Just chatting for free: ChatGPT Free and Go now run on Luna, and Claude Free includes Haiku. For the full consumer picture, see ChatGPT vs Claude vs Gemini vs Grok.
FAQ
When was Claude Haiku 5.5 released?
Anthropic released Claude Haiku 5.5 on October 7, 2026. It is available on the Claude Platform as claude-haiku-5-5 and on Amazon Web Services, Google Cloud and Microsoft Azure.
How much does Claude Haiku 5.5 cost?
For prompts up to 100,000 tokens, $0.10 per million input tokens and $0.50 per million output tokens, with cache reads at $0.01. For prompts over 100,000 tokens, $0.50 input and $2.50 output. The Batch API halves those rates.
Is Haiku 5.5 better than GPT-6 Luna?
On Anthropic's own launch table it beats Luna on computer use, terminal coding, code quality and knowledge work. Artificial Analysis also scores it higher overall but finds it uses about three times as many output tokens at max effort, so Luna can be cheaper per finished task.
Is GPT-6 Luna free in ChatGPT?
Yes. OpenAI says GPT-6 in ChatGPT's Chat tab runs on GPT-6 Luna for Free and Go users, rolling out from October 8, 2026. Paid plans get GPT-6 Sol.
Which is cheaper for long documents?
GPT-6 Luna. It keeps its base price up to 272,000-token prompts, while Haiku 5.5's price rises fivefold above 100,000 tokens. Between those two lengths Luna is five times cheaper per token.
Does Claude Haiku 5.5 replace Haiku 4.5?
Anthropic lists Haiku 5.5 as the current Haiku and Haiku 4.5 as a legacy model. Haiku 5.5 is cheaper per token on short prompts and adds a one-million-token context window, up from 200,000.
