Anthropic Launches Claude Haiku 5.5 With 75% Lower Costs

Anthropic has launched Claude Haiku 5.5, describing it as its fastest, cheapest, and most capable small model to date. The model is available through the Claude Platform, AWS, Google Cloud, and Microsoft Azure.

Like Haiku 4.5, Haiku 5.5 targets high-volume workloads such as summarization, classification, and other large-scale AI tasks.

Anthropic says the new model delivers significant intelligence and speed improvements over Haiku 4.5.

Claude Haiku 5.5 pricing starts at $0.10 per million input tokens

For prompts up to 100,000 tokens, Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens.

For prompts above 100,000 tokens, pricing increases to $0.50 per million input tokens and $2.50 per million output tokens.

Anthropic says this two-tier pricing structure makes Haiku 5.5 around 75% cheaper to run on average than Haiku 4.5.

The lower pricing tier also matches OpenAI GPT-6 Luna on standard per-token pricing.

Haiku 5.5 beats GPT-6 Luna across six listed benchmarks

Anthropic says Claude Haiku 5.5 outperforms GPT-6 Luna across all six benchmarks included in its comparison.

Those tests include GDPval-AA, OSWorld, Terminal-Bench, FrontierCode, and Chartography, with some of the biggest gains appearing in computer use and agentic coding.

Anthropic’s results also show substantial performance improvements over Haiku 4.5.

Independent testing from Artificial Analysis gives Haiku 5.5 an Intelligence Index score of 43, compared with 38 for GPT-6 Luna.

However, the stronger performance comes with higher token usage.

According to Artificial Analysis, Haiku 5.5 uses around 3.5 times more output and reasoning tokens than Luna. That pushes the estimated cost per task to about $0.21, compared with $0.07 for Luna.

Haiku 5.5 generates tokens faster, but its higher token usage means it can take slightly longer to complete a task overall.

Anthropic says Haiku 5.5 improves safety

Anthropic also reports safety improvements in Haiku 5.5.

The company says the model shows fewer instances of misaligned behavior and is less willing than Haiku 4.5 to cooperate with potentially harmful requests.

Claude subscribers are getting monthly API credits

Anthropic is also introducing monthly API credits for some Claude subscribers.

Max 5x subscribers will receive $100 in API credits, while Max 20x users will receive $200. Team subscribers can receive up to $500 in pooled credits.

Users can apply those credits across Anthropic’s models when building applications and AI agents.

The launch follows several other major Anthropic developments. The company recently committed $100 million to Claude Frontier Academy to train 10,000 AI engineers and devoted 80 pages of its IPO prospectus to AI risks.