Anthropic has released Claude Haiku 4.5, the newest addition to its Claude 4 family, offering near-frontier intelligence at one-third the cost and more than twice the speed of its predecessor models. The company described it as the “fastest and most efficient Claude yet,” optimized for real-time, latency-sensitive AI applications.

Five months after Claude Sonnet 4 defined the state of the art, Haiku 4.5 now matches much of its coding and reasoning performance while dramatically improving cost-efficiency. The smaller model even surpasses Sonnet 4 in certain domains, including computer use, making products like Claude for Chrome faster and more capable in everyday contexts.

According to Anthropic’s internal benchmarking, Claude Haiku 4.5 achieves performance levels comparable to larger models — and in some cases rivals GPT-5 — while remaining highly cost-effective.

In software engineering tasks measured by SWE-bench Verified, Haiku 4.5 scored 73.3% accuracy, only four points below Claude Sonnet 4.5 (77.2%), and on par with GPT-5 (72.8%) and GPT-5 Codex (74.5%).

On other evaluations, including t2-bench, OSWorld, and MMMLU, Haiku 4.5 consistently maintained strong mid-to-high 70s and 80s accuracy — outperforming Claude Sonnet 4 and Gemini 2.5 Pro in several categories.

Real-world users are already seeing these benchmark gains translate into tangible performance improvements. Those relying on AI for real-time, low-latency tasks — such as chat assistants, customer service agents, or pair programming — benefit from Haiku 4.5’s blend of high intelligence and rapid response times. Developers using Claude Code also report a markedly more responsive experience, particularly in multi-agent projects and rapid prototyping workflows.

The launch complements Claude Sonnet 4.5, released two weeks earlier, which remains Anthropic’s frontier model and top performer in code generation. Together, they demonstrate the company’s evolving multi-model orchestration strategy.

Claude Haiku 4.5 also enables new forms of model collaboration. Claude Sonnet 4.5, for instance, can analyze a complex problem, divide it into a sequence of subtasks, and direct multiple Claude Haiku 4.5 models to process those subtasks simultaneously, completing the overall task more efficiently.

Anthropic emphasized that Claude Haiku 4.5 is not just faster and cheaper but also safer. The model underwent extensive alignment evaluations, showing the lowest rate of misaligned behaviors across Anthropic’s lineup — outperforming Sonnet 4.5 and Opus 4.1 in automated safety testing.

The model earned an AI Safety Level 2 (ASL-2) classification, reflecting limited risk in sensitive domains such as chemical, biological, radiological, and nuclear (CBRN) knowledge production. By contrast, Sonnet 4.5 and Opus 4.1 remain under the more restrictive ASL-3 designation.

Anthropic detailed its evaluation metrics and reasoning in the Claude Haiku 4.5 system card, emphasizing transparency around safety trade-offs, dataset testing, and policy compliance.

Claude Haiku 4.5 is available everywhere today, including through Claude Code, Claude apps, and third-party integrations on Amazon Bedrock and Google Cloud Vertex AI. The model’s efficiency allows users to accomplish more within existing usage limits while maintaining near-frontier performance.

Developers can access the model directly via the Claude API, using the endpoint claude-haiku-4-5, priced at $1 per million input tokens and $5 per million output tokens — the most economical tier in Anthropic’s current lineup. For complete technical details, benchmark data, and evaluation results, see the Claude Haiku 4.5 model page and official documentation.

Q1: What makes Claude Haiku 4.5 different from Sonnet 4.5?
A: Haiku 4.5 prioritizes speed and cost-efficiency, delivering near-frontier performance for real-time AI tasks at one-third the cost.

Q2: How does Claude Haiku 4.5 perform compared to other models?
A: Benchmarks show it closely matches Claude Sonnet 4.5 and GPT-5 in coding and reasoning tasks, outperforming Gemini 2.5 Pro in multiple benchmarks.

Q3: Where is Haiku 4.5 available?
A: It’s live on the Claude API, Claude apps, Claude Code, Amazon Bedrock, and Google Cloud Vertex AI.

Q4: What is Anthropic’s AI Safety Level classification?
A: ASL-2, indicating limited risk and strong safeguards in sensitive knowledge domains.

Q5: How can developers use Haiku 4.5 and Sonnet 4.5 together?
A: Sonnet 4.5 can orchestrate multiple Haiku 4.5s to perform subtasks in parallel, improving responsiveness and scaling for complex problems.

Claude Haiku 4.5 represents a critical bridge in Anthropic’s tiered AI model strategy — combining affordability, speed, and safety with near-frontier capabilities. For developers and enterprises, it offers a new balance between cost and intelligence, enabling broader deployment of high-quality AI in latency-sensitive environments.

Its ASL-2 safety rating underscores a growing emphasis on responsible model scaling, where smaller, more efficient systems can operate safely without the heavy constraints of frontier models. The introduction of Haiku 4.5 also highlights Anthropic’s ecosystem approach, where models of different sizes collaborate dynamically rather than competing directly.

In the broader AI landscape, Haiku 4.5 signals a pivot from sheer capability toward efficiency, alignment, and orchestration — a shift that could redefine how developers, enterprises, and consumers access advanced AI at scale.

Ultimately, that evolution reflects a maturing AI era — one where progress is measured not by size alone, but by how intelligently, efficiently, and responsibly it is applied.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant used for research and drafting. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.