Claude Haiku 5.5 Launches at $0.10 per Million Input Tokens, the Same List Price as GPT-6 Luna
A.I. / news
Claude Haiku 5.5 Launches at $0.10 per Million Input Tokens, the Same List Price as GPT-6 Luna
Anthropic prices the small model at 90% below Haiku 4.5 for prompts up to 100,000 tokens, then five times higher beyond that line.
Anthropic released Claude Haiku 5.5 on Wednesday at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Those are the same list prices OpenAI set for GPT-6 Luna in September.
The prices, the benchmark table and the model ID claude-haiku-5-5 are on Anthropic's launch page. Anthropic said the model is available now on its own API, Amazon Web Services, Google Cloud and Microsoft Azure.
Haiku 5.5 pricing against Haiku 4.5 and GPT-6 Luna
Anthropic said Haiku 5.5 is priced 90% lower than Haiku 4.5 for requests up to 100,000 tokens. Above that line the model costs $0.50 per million input tokens and $2.50 per million output tokens, which Anthropic described as 50% lower than Haiku 4.5.
| Model | Input per 1M tokens | Output per 1M tokens |
|---|---|---|
| Haiku 5.5, prompts up to 100k | $0.10 | $0.50 |
| Haiku 5.5, prompts over 100k | $0.50 | $2.50 |
| Haiku 4.5 | $1.00 | $5.00 |
| Sonnet 5.5 | $2.00 | $10.00 |
| GPT-6 Luna (iThinkDifferent) | $0.10 | $0.50 |
The Luna figures come from iThinkDifferent, which reported that OpenAI cut Luna from $0.20 to $0.10 input and from $1.20 to $0.50 output on Sept. 23. It called the old figures GPT-5.6 promotional rates. That report mentions no long-prompt tier for Luna.
Anthropic added a footnote that Haiku 5.5 uses a new tokenizer, like Sonnet 5.5 and Opus 5.5, so it uses slightly more tokens per task. A 90% cut in the rate does not become a 90% cut in a bill.
What Anthropic's benchmark table shows
Every score below is vendor-supplied. Anthropic's page does not say which effort setting it used for Luna, and it cites no independent party running either model.
- Sonnet 5.583.9 %
- Haiku 5.572.4 %
- GPT-6 Luna48.9 %
- Haiku 4.515.7 %
Source: Anthropic, Claude Haiku 5.5 launch page, accessed 2026-10-08. Vendor-supplied.
On Terminal-Bench 4.0 the same table gives Haiku 5.5 39.2%, GPT-6 Luna 16.4% and Sonnet 5.5 70.6%. Haiku 4.5 scored 0.0%. Haiku 5.5 trails Sonnet 5.5 on every row Anthropic published, including Humanity's Last Exam at 45.9% without tools against 56.9%.
The page carries six customer quotes. Ze'ev Klapow, Distinguished Software Engineer at HubSpot, said Haiku 5.5 "got the best score we've seen on this suite yet, at 92.8% averaged over three runs." Daniel Campos, Distinguished Engineer at AlphaSense, reported 0.84 against 0.76 for Haiku 4.5 across 400 queries. Anthropic chose which quotes to print.
Context window, limits and complaints
The models overview lists a 1M-token context window, 128K tokens of maximum output and a June 2026 knowledge cutoff. The launch page itself gives no context figure.
Anthropic said Haiku 5.5's cybersecurity safeguards are more restrictive than Haiku 4.5's and still block penetration testing. Its biology safeguards match Sonnet 5 and Opus 5.
On the Hacker News thread, a commenter named minaxir called the 100,000-token cutoff "absurdly low" and said it applies to Haiku but not to Sonnet or Opus. That is one reader's view, not a measurement.
Haiku 5.5 sits in a market where Gemini 4 Argon costs $1.99 a task at promo price and open-weight alternatives such as Qwen3.8-27B ship under Apache 2.0. Anthropic's models page commits to keeping Haiku 5.5 available until at least Oct. 7, 2027.
Sources
More in A.I.
- 01OpenAI Fires Three Safety Staff Nine Days After Publishing Outside-Audit PrinciplesThe company confirmed the dismissals on Oct. 1 but has not said what information moved, who received it, or whether the three first raised concerns internally.
- 02Google's EmbeddingGemma 2 Adds Images, Video and Audio to a 740M EmbedderThe Apache 2.0 weights lift the code-retrieval score by 9.9 points but move the multilingual text score by only 0.21.
- 03Cloudflare's Clef Beats TypeSafe's Jev on Three Tests and Loses on OneThe Apache 2.0 decision model costs nearly six times as much per million tokens as Jev, and its smaller sibling trails badly on one intent benchmark.
- 04Qwen3.8-27B Has 6.78M Downloads and an Apache 2.0 Licence, but Its Card Gives Training Data One LineAlibaba's 27B dense model scores 61.7 on SWE-bench Pro by its own table. The card does not say what it was trained on or what hardware it needs.