Amazon Bedrock Supported Models and Pricing:
- Claude (LLM): Capable of editing, summarizing, and writing text with a max token limit of 100K. On-Demand pricing is $0.01102 per 1000 input tokens and $0.03268 per 1000 output tokens.
- Claude Instant (LLM): Suitable for casual dialogue and document comprehension, also with 100K max tokens. On-Demand pricing is $0.00163 per 1000 input tokens and $0.00551 per 1000 output tokens.
- Command (LLM): For chat, Q&A, and text generation, with a max of 4,096 tokens. On-Demand pricing is $0.0015 per 1000 input tokens and $0.0020 per 1000 output tokens.
- Jurassic-2 Mid and Ultra (LLM): Advanced information extraction with a max of 8,192 tokens. On-Demand pricing is $0.0125 and $0.0188 per 1000 input and output tokens respectively for Mid, and the same pricing for Ultra.
- SDXL0.8 (Stable Diffusion): Image generation with 77 max tokens. On-Demand pricing is $0.018 per image for standard quality and $0.036 per image for premium quality (512x512 or smaller resolution).
- Provisioned Throughput Pricing: Offers specific throughput for consistent workloads. For example, Claude Instant is $39.60 per hour with a 1-month commitment and $22.00 per hour with a 6-month commitment.
- Model Customization: Charges for text generation model customization vary. For instance, Titan Text - Lite customization is $0.0004 per 1000 tokens for training, with a storage cost of $1.95 per month.
Azure OpenAI Supported Models and Pricing:
- Ada (Embeddings): Designed for tasks like anomaly detection, with 8,191 max tokens. Pay-As-You-Go pricing is $0.0001 per 1000 input tokens.
- DALL-E (Image Generation): Generates images from text with 1,000 characters max. Pay-As-You-Go pricing is $2 per 100 images.
- GPT-3.5-Turbo 4k and 16k (LLM): For advanced reasoning and chat, with 4,096 and 16,384 max tokens respectively. Pay-As-You-Go pricing is $0.0015 per 1000 input tokens and $0.002 per 1000 output tokens for 4k, and $0.003 per 1000 input tokens and $0.004 per 1000 output tokens for 16k.
- GPT-4 8k and 32k (LLM): Offers chat functionality with 8,192 and 32,768 max tokens respectively. Pay-As-You-Go pricing is $0.03 per 1000 input tokens and $0.06 per 1000 output tokens for 8k, and $0.06 per 1000 input tokens and $0.12 per 1000 output tokens for 32k.
- Model Customization: Charges for model customization are based on training and hosting time. For example, training for Babbage-002 (GPT Base) costs $34 per compute hour, and hosting is $1.70 per hour.
This detailed comparison of the models and their pricing structures offers insights into the cost-effectiveness and suitability of each model for various applications. Both Amazon Bedrock and Azure OpenAI provide a range of models catering to diverse AI needs, with distinct pricing strategies that cater to different user requirements and budget considerations.
