The "Claude 3" family arrives — a three-tier lineup of Opus, Sonnet, and Haiku
On March 4, 2024, Anthropic announced its new generation of models, the “Claude 3” family. This release introduced, for the first time, a three-tier lineup — the top-performing “Opus,” the balanced “Sonnet,” and the lightweight, fast “Haiku” — establishing the naming and lineup structure that later Claude series releases would follow.
Details
- Three-model lineup: “Opus” aims for maximum capability, “Sonnet” balances speed and intelligence for enterprise workloads, and “Haiku” prioritizes responsiveness as the cheapest option. This release established the practice of choosing a model based on use case
- Benchmarks: Anthropic said Opus outperformed contemporary rival models on major benchmarks including MMLU (undergraduate-level knowledge) and GPQA (graduate-level reasoning). All three models showed improvements in code generation and multilingual support over the previous generation
- Image recognition support: The models can now process images such as photos, graphs, diagrams, and technical documents, enabling them to work with non-text information found in enterprise knowledge bases
- 200K-token context length: All models support 200K tokens, with expansion to over 1 million tokens under consideration for select customers
- Pricing: Opus costs $15/$75 per million input/output tokens, Sonnet $3/$15, and Haiku $0.25/$1.25. Opus and Sonnet were available immediately via the API and claude.ai, with Haiku to follow
What happened next
The Claude 3 family gave Anthropic’s product line a coherent structure all at once, cementing the now-industry-standard approach of offering multiple models for different use cases. Three months later, “Claude 3.5 Sonnet” arrived, surpassing the performance of the top-tier Opus model and rapidly advancing the lineup to its next generation.