Claude Haiku
Anthropic's small + fast tier. Haiku 4.5 (October 2025). $1/M input, $5/M output. Sub-second latency. Use cases: real-time chat, high-volume agent tasks, bulk processing, function-calling routing.
Surprisingly capable for size: Haiku 4.5 matches Sonnet 3.5 (old generation) on many benchmarks. Use cases where Haiku suffices: support chat, simple extraction, classification, summary generation, function-calling routing agents. Anti-use: deep reasoning, complex coding, long multi-step planning — escalate to Sonnet. Competitor: GPT-4o-mini ($0.15/M), Gemini Flash ($0.075/M), Llama 3.3 70B open-weights. Trend 2024-2025: "small models for routine, large models for hard" — agent orchestration with tier routing.