Claude Haiku 4.5 sits within Anthropic's Haiku family of compact language models, designed to balance capability with efficiency for production workloads. On the SWE-bench Verified benchmark, which measures whether a model can resolve real GitHub issues end to end by reading a repository, writing a patch, and passing hidden tests, Claude Haiku 4.5 recorded a score of 73.3 as documented on the benchgraph.dev aggregator. That benchmark draws on 500 tasks across 12 open-source Python repositories, originally published by OpenAI as a human-validated subset of Princeton's SWE-bench, providing a grounded view of agentic coding skill rather than a synthetic leaderboard rank.
The score places Claude Haiku 4.5 in the upper tier of models tracked on SWE-bench Verified, with benchgraph reporting coverage of 112 models on that benchmark and noting saturation as leading models crossed the 70 percent mark during 2025. For practitioners, the model fits workflows where autonomous code repair and repository-grounded reasoning matter, and where a lighter-tier option is desirable compared to the top of the Opus or Claude Mythos Preview lines that score notably higher on the same benchmark. The aggregator's framing emphasizes task-aligned benchmark choice and recent score freshness over headline ranking, which suits Claude Haiku 4.5 as a practical mid-tier choice for agentic coding tasks.
