Anthropic launches Claude 3 family with benchmark improvements and near-instant inference speed, addresses hallucination and refusal rates

Anthropic announced Claude 3 (Opus, Sonnet, Haiku) on March 4, 2024, claiming frontier-level performance on MMLU, GPQA, and GSM8K benchmarks, 2x faster inference than Claude 2/2.1 for Sonnet, 99%+ accuracy on needle-in-haystack recall, and reduced refusal and hallucination rates compared to predecessors. The family is available via API (159 countries), claude.ai, Amazon Bedrock, and Google Cloud Vertex with tiered pricing from $0.25 to $15 per million input tokens. For financial institutions, this establishes a commercially available baseline for enterprise-grade model capabilities and pricing that competitors must meet or beat.