Anthropic’s new Claude Haiku 5.5 has taken the top spot among small AI models in Artificial Analysis’ Intelligence Index, although its high token usage makes the performance lead considerably less impressive in terms of efficiency.
Haiku 5.5 scores 43 at its highest reasoning setting, ahead of GLM-5.3 Flash at 42, Gemini 3.8 Flash at 41, and GPT-6 Luna at 38. It still trails the larger Claude Sonnet 5.5 by 13 points.
Strong Performance Comes With Heavy Token Usage
Artificial Analysis found that Haiku 5.5 consumes around 162,000 output tokens per task at its maximum reasoning setting.
GPT-6 Luna uses roughly 50,000 tokens per task, meaning Haiku can consume more than three times as many tokens.
The difference remains even when performance is matched. At its “high” reasoning setting, Haiku 5.5 scores 38 while using around 55,000 tokens, compared with roughly 50,000 tokens for GPT-6 Luna at the same score.
Artificial Analysis therefore gives GPT-6 Luna an advantage in Pareto efficiency, which measures how much performance a model delivers relative to resources and cost.
Moving Haiku 5.5 from its “xhigh” to “max” setting adds only around two Intelligence Index points while increasing token consumption by roughly 1.8 times.
At maximum effort, Haiku 5.5 reportedly costs around $0.21 per Intelligence Index task, while GPT-6 Luna reaches broadly comparable performance for roughly one-third of that amount.
Haiku Hallucinates Less but Knows Less
Haiku 5.5 performed better in Artificial Analysis’ hallucination testing.
The model hallucinated in around 40% of tested cases, compared with 77% for GPT-6 Luna.
However, GPT-6 Luna showed stronger factual knowledge on AA-Omniscience, scoring 44% accuracy compared with Haiku 5.5’s 36%.
Haiku partly compensates by being more willing to admit when it does not know an answer rather than generating an incorrect response.
Anthropic has also increased the model’s context window significantly, from 200,000 tokens on Haiku 4.5 to one million tokens on Haiku 5.5.
Huge Upgrade Over Haiku 4.5
Anthropic positions Haiku 5.5 as its fastest and cheapest small model for high-volume workloads such as summarization, classification, database queries and customer support.
The model scored 1,620 on GDPval-AA v2.1, compared with 735 for Haiku 4.5.
On Humanity’s Last Exam, Haiku 5.5 reached 45.9% without tools and 57.4% with tools, compared with just 10.2% and 18.7% respectively for its predecessor.
The largest improvement came in computer use. Haiku 5.5 scored 72.4% on the offline subset of OSWorld 2.1, up from 15.7% for Haiku 4.5 and ahead of GPT-6 Luna’s 48.9%.
| Benchmark | Haiku 5.5 | Haiku 4.5 | GPT-6 Luna | Sonnet 5.5 |
|---|---|---|---|---|
| GDPval-AA v2.1 | 1,620 | 735 | 1,437 | 1,840 |
| AA-Briefcase v1.1 | 1,578 | 614 | 1,336 | 1,824 |
| OSWorld 2.1 | 72.4% | 15.7% | 48.9% | 83.9% |
| Terminal-Bench 4.0 | 39.2% | 0% | 16.4% | 70.6% |
| FrontierCode 1.1 | 46.4% | — | 42.4% | 52.1% |
| Chartography | 46.4% | 6.4% | 29.1% | 61.6% |
Anthropic Slashes Haiku Pricing
Haiku 5.5 is also substantially cheaper per token than Haiku 4.5.
For prompts of up to 100,000 tokens, input costs just $0.10 per million tokens, while output costs $0.50 per million.
| Price per 1M Tokens | Haiku 5.5 up to 100K | Haiku 5.5 over 100K | Haiku 4.5 |
|---|---|---|---|
| Cache reads | $0.01 | $0.05 | $0.10 |
| Cache writes | $0.125 | $0.625 | $1.25 |
| Input | $0.10 | $0.50 | $1.00 |
| Output | $0.50 | $2.50 | $5.00 |
Anthropic says this makes Haiku 5.5 around 75% cheaper on average than Haiku 4.5, with savings of up to 90% for shorter prompts.
However, the model uses a new tokenizer and can generate significantly more tokens, meaning real-world savings may be smaller than the headline per-token price reductions suggest.
Adjustable Reasoning Comes to Haiku
Haiku 5.5 is the first model in the Haiku family to offer adjustable reasoning levels, allowing developers to trade higher performance for greater cost and token consumption.
Anthropic recommends it mainly for tightly defined tasks, including summarization, compaction, and sub-agent workloads.
For more demanding agentic coding tasks, the company still recommends Claude Sonnet 5.5 or Opus 5.5.
Haiku 5.5 is available across Anthropic’s platforms as well as Amazon Web Services, Google Cloud and Microsoft Azure.
Anthropic has also cut Sonnet 5.5 cache-read pricing by 50% to $0.10 per million tokens and introduced monthly API credits of up to $500 for some paid subscribers.
The post Claude Haiku 5.5 With 90% Discount is Here appeared first on ProPakistani.
