Anthropic is an artificial intelligence research company founded in 2021 by former OpenAI researchers including Dario Amodei and Paul Christiano. The company focuses on developing safe and reliable AI systems, guided by the principle of “constitutional AI” to ensure the safety and ethics of AI technology.
Company Development:
- 2021: Company founded with substantial investment backing
- 2022: Launched first Claude model
- 2023: Released Claude 2 series, gaining wide recognition
- 2024: Introduced breakthrough Claude 3 series models
Major Model Series:
Claude 3.5 Sonnet:
- Latest generation high-performance model (October and June versions)
- Top performer in quality evaluations
- Supports 200k large context window
- Output speed around 60 tokens/second
- First token response time about 0.8 seconds
- Pricing: $6.00/million tokens
Claude 3 Opus:
- Enterprise-grade premium model
- Excels in complex reasoning tasks
- Supports 200k context window
- Output speed around 27 tokens/second
- First token response time about 2.6 seconds
- Pricing: $30.00/million tokens
Claude 3.5/3 Haiku:
- Lightweight fast-response models
- Claude 3 Haiku: 131 tokens/second
- Claude 3.5 Haiku: 69 tokens/second
- Lowest latency: Claude 3 Haiku at 0.57 seconds
- Cost-effective: Claude 3 Haiku at $0.50/million tokens
Performance Evaluation:
- Leading in MMLU (reasoning and knowledge) tests
- Excellent in GPQA (scientific reasoning) evaluation
- Outstanding in mathematical reasoning (MATH) tests
- Strong performance in code evaluation (HumanEval)
- Top rankings in LMSys Chatbot Arena ELO scores
Innovative Features:
- Full function calling support across all models
- JSON mode output support
- System prompt caching functionality
- Cache saves 90% on input token costs
- Cache Time to Live (TTL) of 5 minutes
Technical Characteristics:
- Employs “constitutional AI” training methodology
- Emphasizes AI safety and ethical guidelines
- Supports multimodal understanding and generation
- Strong contextual understanding capabilities
- Provides stable and reliable API services
Corporate Vision: Anthropic is committed to developing safe, reliable, and ethically conscious AI systems. Through continuous innovation and strict safety standards, the company is driving AI technology toward a more responsible direction. Their Claude series models maintain high performance while particularly emphasizing safety and ethics, providing users with trustworthy AI solutions.
