Uploaded October 2025 | Updated September 2026, 2 hours ago
Claude Haiku 4.5 just blurred the line between “small” and “smart.”
It’s now matching Claude Sonnet 4’s reasoning and coding scores (73% on SWE-bench) at twice the speed and one-third of the cost—proof that distillation works.
Haiku 4.5 isn’t just faster; it’s leaner, running long-context and agentic tasks with impressive 39M-token efficiency. The message is clear: optimization is maturing, and size alone no longer defines capability.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#AInews #Claude45 #Anthropic #short
Claude Haiku 4.5 just blurred the line between “small” and “smart.”
It’s now matching Claude Sonnet 4’s reasoning and coding scores (73% on SWE-bench) at twice the speed and one-third of the cost—proof that distillation works.
Haiku 4.5 isn’t just faster; it’s leaner, running long-context and agentic tasks with impressive 39M-token efficiency. The message is clear: optimization is maturing, and size alone no longer defines capability.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#AInews #Claude45 #Anthropic #short
