Claude Haiku 4.5 Summary

Overview

• Model: Claude Haiku 4.5, Anthropic's latest small model.
• Release Date: October 15, 2025.
• Key Positioning: Delivers near-frontier performance (comparable to the former state-of-the-art Claude Sonnet 4) at significantly lower cost and higher speed.

Key Features & Performance

• Cost & Speed: One-third the cost and more than twice the speed of Claude Sonnet 4.
• Performance: Achieves 90% of Claude Sonnet 4.5's performance in agentic coding evaluations. Surpasses Sonnet 4 on tasks like computer use.
• Use Cases: Ideal for real-time, low-latency applications such as chat assistants, customer service agents, pair programming, and rapid prototyping.
• Orchestration: Can be used with Sonnet 4.5, where the frontier model breaks down complex problems and orchestrates multiple Haiku 4.5 instances to complete subtasks in parallel.

Benchmarks & Quotes

• SWE-bench Verified: Scores 73.3% (averaged over 50 trials).
• Augment Agentic Coding Evaluation: Achieves 90% of Sonnet 4.5's performance.
• Slide Text Generation: Outperforms premium models with 65% accuracy vs. 44%.
• Speed: Runs 4-5 times faster than Sonnet 4.5.

"Claude Haiku 4.5 hit a sweet spot we didn't think was possible: near-frontier coding quality with blazing speed and cost efficiency."

"Speed is the new frontier for AI agents operating in feedback loops. Haiku 4.5 proves you can have both intelligence and rapid output."

Safety Evaluations

• Alignment: Shows a statistically significantly lower overall rate of misaligned behaviors than both Claude Sonnet 4.5 and Claude Opus 4.1, making it Anthropic's safest model by this metric.
• Risk Level: Poses limited risks for CBRN weapons production.
• Safety Standard: Released under AI Safety Level 2 (ASL-2), compared to the more restrictive ASL-3 for Sonnet 4.5 and Opus 4.1.

Availability & Pricing

• Access: Available immediately via the Claude API, Amazon Bedrock, and Google Cloud's Vertex AI.
• API Identifier: claude-haiku-4-5
• Pricing: $1 per million input tokens / $5 per million output tokens.
• Integration: Serves as a drop-in replacement for Haiku 3.5 and Sonnet 4.

Methodology Note (SWE-bench Verified)

"You should use tools as much as possible, ideally more than 100 times. You should also implement your own tests first before attempting the problem."

This prompt addition was used in the reported 73.3% score.