Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- Anthropic launched its new Haiku 4.5 model.
- The small model is faster and more cost-effective.
- Haiku 4.5 is available for all Claude.ai free plans.
While large language models (LLMs) are what we most commonly associate with generative AI products, such as chatbots, small language models (SLMs) have their own perks, including lower costs and higher speeds. Leaning into these advantages, Anthropic is now launching a new small model: Haiku 4.5.
Also: I’ve tested free vs. paid AI coding tools – here’s which one I’d actually use
Despite its smaller size, Haiku 4.5 demonstrated similar levels of coding performance on the SWE-bench verified to Claude Sonnet 4, which, when launched four months ago, Anthropic claimed was the world’s best coding model. The company said Haiku 4.5 can even surpass Claude Sonnet 4 in tasks such as computer usage while being one-third the cost and twice the speed.
What this means for all Claude users
The advanced capabilities, combined with high speed, make it a particularly good fit for real-time, low-latency tasks like those a chatbot can handle, according to the post. All users will be able to put Haiku 4.5 to the test, as the model is available in Claude apps, including the Claude.ai chatbot.
Beyond coding capabilities, the model performed competitively against Claude Sonnet 4.5, the company’s flagship LLM that launched at the end of September, on a series of benchmarks, including the MMMU, which tests for visual reasoning; AIME 2025, which tests for high school-level math; and r2-bench, which evaluates for an agentic tool user.
On the safety front, Anthropic said that Haiku 4.5 has shown low rates of concerning behaviors and was “substantially more aligned” than its predecessor, Claude Haiku 3.5, and “significantly” more aligned than Claude Sonnet 4.5 and Claude Opus 4.1, making it the company’s safest model yet. Its performance across safety evaluations earned it an AI Safety Level 2 standard, which is internal to Anthropic, not determined by a third-party tester or external organization. The company offered more details on the designation in the model’s system card.
Developer perks
Beyond enjoying a cheaper yet still capable model, developers will also be able to use Haiku 4.5 in combination with the heavier Claude Sonnet 4.5, which Anthropic claimed is still the most powerful coding model available.
Also: Even the best AI agents are thwarted by this protocol – what can be done
Sonnet 4.5 can be used to break down a problem a develop a multi-step plan, which a “team of multiple Haiku 4.5s” could then complete task by task, Anthropic wrote.
Want more stories about AI? Sign up for AI Leaderboardour weekly newsletter.
Claude Haiku 4.5 is available for developers on the API, Amazon Bedrock, and Google Cloud’s Vertex AI. Anthropic said pricing is now $1/$5 per million input and output tokens.
