Anthropic releases efficient Claude Haiku 4.5 AI model

Anthropic has launched Claude Haiku 4.5, a compact AI model that matches the coding performance of its larger counterpart from five months ago, but at one-third the cost and over twice the speed. Available immediately to users of the Claude app, web, and API, the model targets real-time tasks like coding assistance. Benchmarks show it scoring highly on coding tests, approaching levels of advanced models like GPT-5.

On October 15, 2025, Anthropic introduced Claude Haiku 4.5, the latest in its small-model lineup within the Claude family, which includes Haiku for smaller tasks, Sonnet as medium-sized, and Opus as large. This new version uses distillation techniques to replicate capabilities of older, larger models like Claude Sonnet 4—released five months earlier—particularly in functional areas such as coding, while sacrificing some stored knowledge for broader analysis topics like foreign policy or history.

Performance-wise, Haiku 4.5 achieved 73.3 percent on the SWE-bench Verified coding benchmark, slightly edging out Sonnet 4's 72.7 percent and reportedly surpassing it in tasks like computer usage. Anthropic claims it nears the coding scores of OpenAI's GPT-5, though these are self-reported results that warrant independent verification. For context, Anthropic's current frontier model, Claude Sonnet 4.5 from late September 2025, remains the company's top coding option.

Cost savings are significant: via API, Haiku 4.5 charges $1 per million input tokens and $5 per million output tokens, compared to Sonnet 4.5's $3 input and $15 output, and Opus 4.1's $15 input and $75 output. It's included in Claude web and app subscriptions and serves as a drop-in replacement for Haiku 3.5 and Sonnet 4.

"Users who rely on AI for real-time, low-latency tasks like chat assistants, customer service agents, or pair programming will appreciate Haiku 4.5’s combination of high intelligence and remarkable speed," Anthropic stated. The model integrates into multi-model setups, where Sonnet 4.5 could plan complex problems and delegate subtasks to multiple Haiku 4.5 instances for parallel processing, enhancing efficiency in agentic coding like Claude Code. Anthropic has provided a system card and developer documentation for further details.

यह वेबसाइट कुकीज़ का उपयोग करती है

हम अपनी साइट को बेहतर बनाने के लिए एनालिटिक्स के लिए कुकीज़ का उपयोग करते हैं। अधिक जानकारी के लिए हमारी गोपनीयता नीति पढ़ें।
अस्वीकार करें