Anthropic Launches Claude Sonnet 4.5 & Haiku 4.5
Claude Sonnet 4.5 Launch
Anthropic unveiled Claude Sonnet 4.5 on September 28, 2025, making a bold claim as the world's best coding model based on industry benchmarks like SWE-bench Verified.
The most impressive detail from their announcement wasn't the benchmarks—it was watching the model code autonomously for up to 30 hours during enterprise trials. Developers saw it build entire applications, set up database services, purchase domain names, and even perform SOC 2 security audits with minimal human intervention.
Anthropic kept the pricing identical to Claude Sonnet 4 at $3 per million input tokens and $15 per million output tokens, making this a direct upgrade against competitors. The company also emphasized improvements in AI alignment, with lower rates of sycophancy and deception compared to previous models, plus better resistance to prompt injection attacks.
You can access Sonnet 4.5 through the Claude API, their official chat interface, or through multi-modal platforms like Perplexity.
Claude Haiku 4.5: The Budget Performance King
Just weeks later in mid-October, Anthropic launched Claude Haiku 4.5, and it's already disrupting pricing expectations across the AI model landscape.
Despite being positioned as Anthropic's compact model, Haiku 4.5 matches Claude Sonnet 4's performance on coding tasks while starting at just $1 per million input tokens and $5 per million output tokens. On the SWE-bench Verified coding benchmark, Haiku 4.5 scored 73.3% — putting it on par with much larger models like Sonnet 4 and OpenAI's GPT-5.
The model also demonstrates superior computer usage capabilities compared to its predecessor, making it particularly valuable for autonomous agent workflows that require tool use and API interactions.
With additional savings of up to 90% through prompt caching and 50% via Message Batches API, I think Haiku 4.5 has the potential to make enterprise-scale AI deployment economically viable for startups and small/mid-sized companies. For developers building customer support bots, content automation pipelines, or code review systems, Haiku 4.5 offers the sweet spot of performance, speed, and cost.
Read more from the original sources: