Introducing the next generation of Claude \ Anthropic
Claude 3 models offer near-human intelligence, 2x faster Sonnet, and vision capabilities.
Anthropic unveiled the Claude 3 family on March 4, 2024, setting new industry benchmarks across cognitive tasks. The three models—Haiku, Sonnet, and Opus—offer escalating intelligence, speed, and cost options. Opus achieves near-human comprehension on complex tasks, outpacing competitors on undergraduate (MMLU), graduate (GPQA), and math (GSM8K) benchmarks. Sonnet delivers 2x faster responses than Claude 2.1 while improving accuracy. Haiku, the fastest and most cost-effective, can read a 10k-token research paper in under 3 seconds.
Vision capabilities enable processing photos, charts, and diagrams, with enterprise customers seeing up to 50% of knowledge bases in visual formats. The models also show a two-fold accuracy improvement over Claude 2.1 on challenging factual questions, with fewer unnecessary refusals and upcoming citation features. All models start with a 200K context window, with potential 1M-token support for select customers. Opus, Sonnet, and Haiku are available now via claude.ai and API in 159 countries.
- Claude 3 Opus outperforms peers on MMLU, GPQA, and GSM8K benchmarks with near-human comprehension.
- Sonnet is 2x faster than Claude 2.1; Haiku reads 10k tokens in under 3 seconds at minimal cost.
- All models feature vision (photos, charts), 200K context, fewer refusals, and two-fold accuracy improvement.
Why It Matters
Claude 3 provides enterprises with faster, more accurate AI at tiered costs, enabling real-time automation and complex analysis.