Speed Meets Intelligence

Anthropic today releases Claude Haiku 4.5, the latest addition to the Claude model family. This small model delivers near-frontier coding performance while maintaining exceptional speed and cost-efficiency—a breakthrough in AI model economics.

What was recently at the frontier is now cheaper and faster. Five months ago, Claude Sonnet 4 was state-of-the-art. Today, Claude Haiku 4.5 provides similar coding performance but at one-third the cost and more than twice the speed.

Remarkably, Claude Haiku 4.5 even surpasses Claude Sonnet 4 on certain tasks, including computer use capabilities. These advances make applications like Claude for Chrome faster and more useful than ever before.

The Perfect Balance: Intelligence, Speed, and Cost

For users relying on AI for real-time, low-latency tasks, Claude Haiku 4.5 offers a compelling option:

- Chat assistants: Near-instant responses with sophisticated reasoning
- Customer service agents: Quick resolution with contextual understanding
- Pair programming: Responsive coding assistance from multiple Haikus orchestrated by Sonnet 4.5
- Rapid prototyping: Speed advantage enables faster iteration cycles

Claude Code users will notice significantly improved responsiveness, whether working on single-agent projects or orchestrating multiple Haiku agents in parallel.

Benchmarking Results

Claude Haiku 4.5 achieves 73.3% on SWE-bench Verified using extended thinking (128K budget), demonstrating remarkable coding capabilities for a small model. In Augment's agentic coding evaluation, it reaches 90% of Sonnet 4.5 performance—a performance level that would have been state-of-the-art just six months ago.

The model runs 4-5 times faster than Sonnet 4.5 while costing a fraction of the price. This speed advantage opens entirely new use cases for agentic AI systems operating in feedback loops.

Multi-Agent Orchestration

Claude Haiku 4.5 enables a new development pattern: Sonnet 4.5 can break down complex problems into multi-step plans, then orchestrate a team of multiple Haiku 4.5 agents to complete subtasks in parallel. This approach combines frontier reasoning with sub-second response times.

Safety and Alignment

Claude Haiku 4.5 underwent extensive safety and alignment evaluations. The model showed:

- Low concerning behaviors: Substantially more aligned than predecessor Claude Haiku 3.5
- Safest by metric: Lower overall misaligned behavior rate than both Sonnet 4.5 and Opus 4.1 in automated assessments
- CBRN risk: Only limited risks for chemical, biological, radiological, and nuclear weapons production
- ASL-2 classification: Released under the less restrictive AI Safety Level 2 standard

Detailed safety testing results are available in the Claude Haiku 4.5 system card.

Availability and Pricing

Claude Haiku 4.5 is available immediately:

- Claude Code & Apps: Available to all users
- API: Use claude-haiku-4-5 via Claude API - Cloud platforms: Amazon Bedrock and Google Cloud Vertex AI - Pricing: $1/$5 per million input and output tokens

Early Feedback

Early partners report impressive results:

- Augment: 90% of Sonnet 4.5 performance on agentic coding with exceptional speed
- Warp: Excellent sub-agent orchestration and computer use, making AI-assisted development feel instantaneous
- GitHub Copilot: Efficient code generation with comparable quality to Sonnet 4 but at faster speeds
- Slide generation: 65% accuracy on instruction-following vs. 44% from premium tier models—game-changing for unit economics

TL;DR

- Claude Haiku 4.5 delivers 90% of Sonnet 4.5 coding performance at 1/3 the cost
- 4-5x faster than Sonnet 4.5 with 73.3% SWE-bench Verified performance (128K thinking)
- Outperforms Sonnet 4 on some tasks like computer use and instruction-following
- Safest model by automated alignment metrics, released under ASL-2
- Multi-agent orchestration: Perfect for Sonnet 4.5 to coordinate multiple Haikus in parallel


Source: Anthropic Blog: Introducing Claude Haiku 4.5