- enterprise-h2ogpte/rag_benchmark/results/test_client_e2e.md at main · h2oai/enterprise-h2ogpte
- Client Code Examples, Use Cases and Benchmarks for Enterprise h2oGPTe RAG-Based GenAI Platform - h2oai/enterprise-h2ogpte
Anthropic's popular Claude 3 is now available on GCP.
(Actually, it seems to have been available for a while.)
It's not fully available yet, only Sonnet and Haiku are accessible, and Opus is still [Coming Soon].
First, here are the metrics evaluated by H2O.ai using RAG:
Evaluation Results by LLM
Source: https://github.com/h2oai/enterprise-h2ogpte/blob/main/rag_benchmark/results/test_client_e2e.md
Here's a comparison table with Gemini, which I personally prefer.
Pricing and RAG Accuracy of LLM Models Available on GCP
The rightmost table is a selection of 5 LLM models from the above table.
The price is for input and output for 1 million tokens each.
Looking solely at tokens, Claude 3 Haiku seems to be the cheapest currently. (Actually, Gemini Pro wasn't an expensive option either...)
It would be beneficial to mix and match these models based on your needs.
Comments0