Claude's Latest Model Is Cheaper And Faster Than Sonnet 4 - And Free

Anthropic

Follow ZDNET: Add america arsenic a preferred source on Google.

ZDNET's cardinal takeaways

• Anthropic launched its caller Haiku 4.5 model.
• The mini exemplary is faster and much cost-effective. 
• Haiku 4.5 is disposable for each Claude.ai free plans. 

While ample connection models (LLMs) are what we astir commonly subordinate pinch generative AI products, specified arsenic chatbots, mini connection models (SLMs) person their ain perks, including little costs and higher speeds. Leaning into these advantages, Anthropic is now launching a caller mini model: Haiku 4.5.  

Also: I've tested free vs. paid AI coding devices - here's which 1 I'd really use

Despite its smaller size, Haiku 4.5 demonstrated akin levels of coding capacity connected nan SWE-bench verified to Claude Sonnet 4, which, erstwhile launched 4 months ago, Anthropic claimed was nan world's champion coding model. The institution said Haiku 4.5 tin moreover surpass Claude Sonnet 4 successful tasks specified arsenic machine usage while being one-third nan costs and doubly nan speed. 

What this intends for each Claude users

The precocious capabilities, mixed pinch precocious speed, make it a peculiarly bully fresh for real-time, low-latency tasks for illustration those a chatbot tin handle, according to nan post. All users will beryllium capable to put Haiku 4.5 to nan test, arsenic nan exemplary is disposable successful Claude apps, including nan Claude.ai chatbot. 

Beyond coding capabilities, nan exemplary performed competitively against Claude Sonnet 4.5, nan company's flagship LLM that launched astatine nan extremity of September, connected a bid of benchmarks, including nan MMMU, which tests for ocular reasoning; AIME 2025, which tests for precocious school-level math; and r2-bench, which evaluates for an agentic instrumentality user. 

Anthropic

On nan information front, Anthropic said that Haiku 4.5 has shown debased rates of concerning behaviors and was "substantially much aligned" than its predecessor, Claude Haiku 3.5, and "significantly" much aligned than Claude Sonnet 4.5 and Claude Opus 4.1, making it nan company's safest exemplary yet. Its capacity crossed information evaluations earned it an AI Safety Level 2 standard, which is soul to Anthropic, not wished by a third-party tester aliases outer organization. The institution offered much specifications connected nan nickname successful nan model's strategy card. 

Developer perks 

Beyond enjoying a cheaper yet still tin model, developers will besides beryllium capable to usage Haiku 4.5 successful operation pinch nan heavier Claude Sonnet 4.5, which Anthropic claimed is still nan astir powerful coding exemplary available. 

Also: Even nan champion AI agents are thwarted by this protocol - what tin beryllium done

Sonnet 4.5 tin beryllium utilized to break down a problem a create a multi-step plan, which a "team of aggregate Haiku 4.5s" could past complete task by task, Anthropic wrote. 

Want much stories astir AI? Sign up for AI Leaderboard, our play newsletter.

Claude Haiku 4.5 is disposable for developers connected nan API, Amazon Bedrock, and Google Cloud's Vertex AI. Anthropic said pricing is now $1/$5 per cardinal input and output tokens.