Anthropic releases Claude Haiku 5.5 with 75% cost reduction and new effort settings

Anthropic has released Claude Haiku 5.5, positioning it as the cheapest, fastest, and most capable small model in its lineup. Designed for high-volume, cost-sensitive tasks such as summaries, database queries, and classification, it also serves effectively as a subagent for coding work alongside Opus 5.5 and Sonnet 5.5. The model is priced approximately 75% lower than its predecessor, Haiku 4.5, on average, with specific reductions of 90% for requests up to 100,000 tokens.

Performance and Benchmarks

Haiku 5.5 demonstrates significant improvements over Haiku 4.5 across various benchmarks, often approaching or exceeding the performance of larger models in specific domains. Key metrics include:

• Knowledge Work: Scored 1620 on GDPval-AA v2.1 (vs. 735 for Haiku 4.5) and 1578 on AA-Briefcase v1.1 (vs. 614 for Haiku 4.5).
• Computer Use: Achieved 72.4% on OSWorld 2.1 (offline subset), a substantial jump from Haiku 4.5's 15.7%.
• Reasoning: Scored 45.9% on Humanity’s Last Exam (no tools) and 57.4% with tools, compared to 10.2% and 18.7% respectively for Haiku 4.5.
• Agentic Coding: Recorded 39.2% on Terminal-Bench 4.0 and 46.4% on FrontierCode 1.1 (Main).

Early customer testing corroborates these gains. Asana reported a 30% reduction in latency and up to 2.5x faster inference per agent turn. HubSpot noted a 92.8% average score on CRM tasks, the highest in their suite, with the lowest false positive rate. AlphaSense observed a statistically significant improvement in query accuracy (0.84 vs. 0.76) for document-based questions.

New Features and Pricing Adjustments

Haiku 5.5 is the first Haiku-class model to feature an adjustable effort setting, allowing users to balance cost against intelligence. Pricing for Haiku 5.5 is structured as follows (per 1 million tokens):

• Input tokens: $0.10 (prompts up to 100k) / $0.50 (over 100k)
• Output tokens: $0.50 (prompts up to 100k) / $2.50 (over 100k)
• Cache reads: $0.01 (prompts up to 100k) / $0.05 (over 100k)

Alongside this launch, Anthropic is halving the price of cache reads for Claude Sonnet 5.5 to $0.10 per million tokens, reducing overall costs for agentic work by approximately 20%. Additionally, Max and Team subscribers will receive monthly API credits ($100 for Max 5x, $200 for Max 20x, and up to $500 pooled for Team) to support application development.

Availability and Safety

Claude Haiku 5.5 is available immediately on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure, via the model ID claude-haiku-5-5. Safety evaluations indicate major improvements in alignment, with fewer instances of misaligned behavior. Cybersecurity safeguards are more restrictive than Haiku 4.5 but permit a wider range of defensive tasks than Sonnet 5.5, while still blocking penetration testing. Biology safeguards remain consistent with Sonnet 5 and Opus 5.