Anthropic’s new Claude Haiku 5.5 costs 10 cents per million input tokens and 50 cents per million output tokens. That’s a tenth of what Haiku 4.5 charged, and it’s exactly the price OpenAI puts on GPT-6 Luna, the model Anthropic chose to benchmark against.
Claude Haiku 5.5, released on October 7, is the kind of model most people never pick by name but that developers call millions of times a day: for summaries, classification, support chatbots and “subagents” that do small jobs for a bigger model. At that volume, a few cents per million tokens decides which company gets the business. The headline price comes with conditions, though, and the benchmark win is Anthropic’s own scorecard.
Claude Haiku 5.5’s price has a 100,000-token cliff
The 10-cent rate applies only to prompts up to 100,000 tokens. Above that, Anthropic charges five times as much: $0.50 in and $2.50 out. Anthropic says prompts of 100,000 tokens or less made up about 90% of requests to its previous Haiku model, so most calls should land in the cheap tier. GPT-6 Luna has a threshold too, but a higher one: OpenAI’s model page says prompts over 272,000 input tokens are billed at twice the input rate and 1.5 times the output rate.
| Per million tokens | Claude Haiku 5.5 (prompts up to / over 100K) | Claude Haiku 4.5 | Claude Sonnet 5.5 | GPT-6 Luna (prompts up to 272K) |
|---|---|---|---|---|
| Input | $0.10 / $0.50 | $1.00 | $2.00 | $0.10 |
| Output | $0.50 / $2.50 | $5.00 | $10.00 | $0.50 |
| Cached input (reads) | $0.01 / $0.05 | $0.10 | $0.10 | $0.01 |
Prices are from Anthropic’s model documentation and OpenAI’s Luna page, checked October 9. Both companies discount batch jobs by 50%.
Here’s the fine print that matters most. Anthropic’s docs say Haiku 5.5 uses a newer tokenizer, so the same text counts as roughly 30% more tokens than on Haiku 4.5. That’s why Anthropic’s own savings estimate is “around 75% less to run” on average, not 90%. It’s also why a token-for-token match with Luna doesn’t mean equal bills: the two companies count text differently, and the cost of finishing a task depends on how many tokens each model burns along the way.
Anthropic’s benchmarks put Haiku 5.5 ahead of GPT-6 Luna
Anthropic’s launch table compares Haiku 5.5 with Haiku 4.5, GPT-6 Luna and, for reference, its mid-size Sonnet 5.5. These are Anthropic’s runs, and the table doesn’t say which effort setting it used for Luna.
| Benchmark (Anthropic’s results) | Haiku 5.5 | Haiku 4.5 | GPT-6 Luna | Sonnet 5.5 |
|---|---|---|---|---|
| GDPval-AA v2.1 (knowledge work, score) | 1620 | 735 | 1437 | 1840 |
| OSWorld 2.1, offline subset (computer use) | 72.4% | 15.7% | 48.9% | 83.9% |
| Terminal-Bench 4.0 (agentic coding) | 39.2% | 0.0% | 16.4% | 70.6% |
| FrontierCode 1.1 Main (agentic coding) | 46.4% | not listed | 42.4% | 52.1% |
| Humanity’s Last Exam, no tools | 45.9% | 10.2% | not listed | 56.9% |
The jump over Haiku 4.5 is the striking part: on Anthropic’s computer-use test the score goes from 15.7% to 72.4%. Against Luna, Anthropic shows a clear lead on computer use and Terminal-Bench and a narrower one on FrontierCode. OpenAI’s own Luna numbers, from its September launch post, use different tests and settings, so the two sets can’t be lined up directly.
Anthropic also published customer results. HubSpot’s Ze’ev Klapow, a distinguished software engineer, said Haiku 5.5 “got the best score we’ve seen on this suite yet, at 92.8% averaged over three runs” on the company’s internal CRM tests. And Anthropic is candid about the limits: it says Sonnet 5.5 and Opus 5.5 “remain better choices for complex agentic coding tasks,” and that Haiku 5.5 suits narrower jobs like summarization and subagent work.
A 1-million-token window and an effort dial
According to Anthropic’s documentation, Haiku 5.5 has a 1-million-token context window, writes up to 128,000 tokens per response (300,000 through the batch API with a beta header), takes text and images as input and has a reliable knowledge cutoff of June 2026. It’s the first Haiku with an effort setting, which lets developers trade cost for more thinking; the default is medium. One gotcha for anyone swapping it in: setting temperature, top_p or top_k to anything but the default returns an error.
Developers call it as claude-haiku-5-5 on Anthropic’s API, Amazon Bedrock, Google Cloud and Microsoft Foundry. On Claude.ai, Anthropic says Free, Pro, Max, Team and Enterprise users can select it, and it’s in Claude Code. Anthropic says it’s its fastest model to date at standard speed, but it hasn’t published a tokens-per-second figure.
Sonnet got cheaper too, and Max subscribers get API credit
Anthropic used the launch to cut elsewhere. Cache reads on Sonnet 5.5 drop from $0.20 to $0.10 per million tokens, which the company estimates makes Sonnet about 20% cheaper on most agentic work. And this week Anthropic is adding monthly API credits for subscribers: $100 for Max 5x, $200 for Max 20x and up to $500 for Team plans, pooled across users.
On safety, Anthropic says Haiku 5.5’s cybersecurity safeguards are stricter than Haiku 4.5’s but looser than those on its other recent models, permitting a wider range of defensive tasks than Sonnet 5.5’s safeguards while still blocking penetration testing. Its system card has the details.
The next test is independent. Until third-party evaluators run Haiku 5.5 and GPT-6 Luna side by side at matched settings, the head-to-head rests on Anthropic’s table. OpenAI, meanwhile, has just put Luna in front of every free ChatGPT user, which we covered in our report on GPT-6 coming to the free tier. For now, the floor of the market sits at a dime per million tokens, and the next move belongs to whoever publishes cost-per-task numbers that someone else can check.
Frequently asked questions
How much does Claude Haiku 5.5 cost?
On Anthropic’s API, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Longer prompts cost $0.50 and $2.50. Batch requests are half price, and cached input reads start at $0.01 per million tokens.
Is Claude Haiku 5.5 free on Claude.ai?
Anthropic says Free, Pro, Max, Team and Enterprise users can all select Haiku 5.5 on Claude.ai, on the web and in the iOS and Android apps. Usage limits still depend on your plan.
Is Claude Haiku 5.5 cheaper than GPT-6 Luna?
The list prices are identical for typical prompts: $0.10 in and $0.50 out per million tokens. Real costs can differ because the companies use different tokenizers, apply long-prompt surcharges at different thresholds (100,000 tokens for Haiku, 272,000 for Luna) and models use different numbers of tokens to finish the same task.
What is Claude Haiku 5.5’s context window?
One million tokens, with up to 128,000 output tokens per response. Anthropic lists its reliable knowledge cutoff as June 2026.
Sources
All accessed October 9, 2026.
- Anthropic, “Claude Haiku 5.5” announcement, benchmark and pricing tables (Oct 7, 2026): anthropic.com
- Anthropic, Claude Haiku product page (availability on Claude.ai and in Claude Code): anthropic.com
- Anthropic, Claude Haiku 5.5 model documentation (specs, pricing, tokenizer note): platform.claude.com
- Anthropic on X, launch thread (Oct 7, 2026): x.com
- OpenAI API docs, GPT-6 Luna model page and pricing: developers.openai.com
- OpenAI, “Introducing GPT-6 Sol and Luna” (September 2026): openai.com
- AWS Machine Learning Blog, “Introducing Claude Haiku 5.5 on AWS” (Oct 7, 2026): aws.amazon.com
- TechRepublic, “Anthropic Launches Claude Haiku 5.5 With API Prices Starting at $0.10” (Oct 2026): techrepublic.com
Related: GPT-6 is free in ChatGPT, but it’s Luna · What is a robot foundation model? · More LLM news
Last updated: October 9, 2026. To report an error, see our corrections page. Articles are drafted with AI assistance and reviewed and edited by an editor; see our editorial policy.
