Claude Haiku 5.5
Photo by Zdeněk Macháček on Unsplash
New Haiku model just dropped: Haiku 5.5, released 7 October 2026 (Anthropic, 2026).
Feels like a lifetime since Anthropic updated Haiku. Turns out Haiku 4.5 came out on 15 October 2025, almost exactly a year ago - which is practically a lifetime in AI terms (Anthropic, 2025).
Clearly priced to compete directly with GPT-6 Luna, which is a great, cheap model that's been my go-to when I need a cost-efficient option.
The per-token costs are roughly on par with Luna, although they get 5x more expensive after 100k tokens, which is interesting.
Specs Haiku vs Luna
| Specification | Claude Haiku 5.5 | GPT-6 Luna |
|---|---|---|
| API identifier | claude-haiku-5-5 |
gpt-6-luna |
| Input / output | Text and images / text | Text and images / text |
| Context window | 1M tokens | 1,050,000 tokens (922,000 max input) |
| Maximum output | 128K tokens | 128K tokens |
| Knowledge cutoff | June 2026 | 18 May 2026 |
| Reasoning | Adaptive thinking, on by default | Reasoning effort, including none |
| Effort levels | low to max, default medium |
none to max, default medium |
As of 8 October 2026 (Anthropic) (OpenAI).
Pricing and Cache Settings
Haiku's price per token goes up 5x over 100K. Luna also has an increase for larger context sizes, but only 2x/1.5x input/output, and its threshold is 272K input tokens. On both, a prompt over the threshold pays the higher rate for the whole request (Anthropic) (OpenAI).
Prices in USD per million tokens, as of 8 October 2026.
| Usage | Haiku 5.5, up to 100K | Haiku 5.5, over 100K | Luna, up to 272K | Luna, over 272K |
|---|---|---|---|---|
| Input, uncached | $0.10 | $0.50 | $0.10 | $0.20 |
| Input, cache read | $0.01 | $0.05 | $0.01 | $0.02 |
| Cache write | $0.125 (5 min), $0.20 (1 hour) | $0.625 (5 min), $1.00 (1 hour) | $0.125 | $0.25 |
| Output, including reasoning | $0.50 | $2.50 | $0.50 | $0.75 |
Also, of note, the same launch halved Claude Sonnet 5.5's cache-read price, from $0.20 to $0.10 per million tokens (Anthropic, 2026).
Tokenisers
Haiku 5.5 uses the tokeniser introduced with Claude Opus 4.7, which tends to produce about 30% more tokens than Haiku 4.5 (and other models before Opus 4.7) for the same text (Anthropic).
I counted six files with each provider's free token-counting endpoint (Anthropic's count_tokens and OpenAI's responses/input_tokens) on 8 October 2026:
| Sample | Haiku 5.5 tokens | Luna tokens | Ratio |
|---|---|---|---|
| Survey summary (prose) | 9,992 | 5,966 | 1.67 |
| AudioLM summary (prose) | 8,024 | 4,890 | 1.64 |
| Ansible tutorial (prose and commands) | 6,085 | 4,066 | 1.50 |
| GPT 6 Sol and Luna article (prose and code) | 13,847 | 8,946 | 1.55 |
| Python rendering script | 15,708 | 10,037 | 1.56 |
| Generated HTML game | 11,393 | 7,630 | 1.49 |
So about 1.5 to 1.7 times as many tokens as Luna, which means Haiku costs about 1.5 to 1.7 times as much for the same text.
Benchmarks
Anthropic's announcement compares Haiku 5.5 with GPT-6 Luna, and as you might expect, they report a few improvements over Luna, some of them pushing towards Sonnet 5.5 quality (Anthropic, 2026).
| Benchmark | Haiku 5.5 | GPT-6 Luna | Haiku 4.5 | Sonnet 5.5 |
|---|---|---|---|---|
| GDPval-AA v2.1, Elo | 1620 | 1437 | 735 | 1840 |
| OSWorld 2.1, offline subset | 72.4% | 48.9% | 15.7% | 83.9% |
| Terminal-Bench 4.0 | 39.2% | 16.4% | 0.0% | 70.6% |
| FrontierCode 1.1, Main | 46.4% | 42.4% | n/a | 52.1% |
Overall, Luna is still the cheaper model, but potentially the performance gains from Haiku might be worth the extra token cost.
References
Anthropic. Pricing. Living documentation. Accessed 8 October 2026. URL: https://platform.claude.com/docs/en/about-claude/pricing (visited on 2026-10-08). ↩
Anthropic. Claude Haiku 5.5. Living documentation. Accessed 8 October 2026. URL: https://platform.claude.com/docs/en/models/haiku-5-5/overview (visited on 2026-10-08). ↩ 1 2
Anthropic. Introducing Claude Haiku 4.5. 2025. Published 15 October 2025. Accessed 8 October 2026. URL: https://www.anthropic.com/news/claude-haiku-4-5 (visited on 2026-10-08). ↩
Anthropic. Introducing Claude Haiku 5.5. 2026. Published 7 October 2026. Accessed 8 October 2026. URL: https://www.anthropic.com/claude-haiku-5-5 (visited on 2026-10-08). ↩ 1 2 3
OpenAI. Pricing. Living documentation. Accessed 8 October 2026. URL: https://developers.openai.com/api/docs/pricing (visited on 2026-10-08). ↩
OpenAI. GPT-6 Luna. Living documentation. Accessed 8 October 2026. URL: https://developers.openai.com/api/docs/models/gpt-6-luna (visited on 2026-10-08). ↩
Comments
Reply to this post on Bluesky or Mastodon to join the conversation.