Understand and compare
Gemini Ultra
vs.
GPT-4
Try
Shepherd
Make it trivial for customers to self-host your SaaS.
Overview
Gemini Ultra
|
GPT-4
|
|
|---|---|---|
|
Provider
The entity that provides this model.
|
Google
|
OpenAI
|
|
Input Context Window
The number of tokens supported by the input context window.
|
32.8K
characters
|
8,192
tokens
|
|
Maximum Output Tokens
The number of tokens that can be generated by the model in a single request.
|
8,192
characters
|
8,192
tokens
|
|
Release Date
When the model was first released.
|
Unknown
|
2023-03-14
|
|
Knowledge Cutoff
Limit on the knowledge base used by the model.
|
Unknown
|
September 2021
|
|
Open Source
|
|
|
|
API Providers
The providers that offer this model. (This is not an exhaustive list.)
|
|
|
Pricing
|
Gemini Ultra
|
GPT-4
|
|
|---|---|---|
|
Input
Cost of input data provided to the model.
|
Pricing not available.
|
$30.00
per million tokens
|
|
Output
Cost of output tokens generated by the model.
|
Pricing not available.
|
$60.00
per million tokens
|
Benchmarks
Compare relevant benchmarks between Gemini Ultra
and GPT-4.
|
Gemini Ultra
|
GPT-4
|
|
|---|---|---|
|
MMLU
Evaluating LLM knowledge acquisition in zero-shot and few-shot settings.
|
83.7
(5-shot)
|
86.4
(5-shot)
|
|
MMMU
A wide ranging multi-discipline and multimodal benchmark.
|
59.4
(0-shot pass@1)
|
34.9
|
|
HellaSwag
A challenging sentence completion benchmark.
|
Benchmark not available.
|
95.3
(10-shot)
|
|
GSM8K
Grade-school math problems benchmark.
|
88.9
(11-shot)
|
92.0
(5-shot)
|
|
HumanEval
A benchmark to measure functional correctness for synthesizing programs from docstrings.
|
74.4
(0-shot)
|
67.0
(0-shot)
|
|
MATH
Benchmark performance on Math problems ranging across 5 levels of difficulty and 7 sub-disciplines.
|
53.2
(4-shot Minerva Prompt)
|
Benchmark not available.
|
Gemini Ultra, developed by Google, features a large context window of 32768 tokens. The model has excelled in benchmarks like MMMU with a score of 59.4 in a 0-shot pass@1 scenario and MMLU with a score of 83.7 in a 5-shot scenario.
GPT-4, developed by OpenAI, features a context window of 8192 tokens. The model costs 3.0 cents per thousand tokens for input and 6.0 cents per thousand tokens for output. It was released on March 14, 2023, and has achieved impressive scores in benchmarks like HellaSwag with a score of 95.3 in a 10-shot scenario and MMLU with a score of 86.4 in a 5-shot scenario.
Llama 3.1 70B Instruct
Mistral 7B Instruct
Measure & Improve LLM
Product Performance.
Get Started
© 2023-2024 Context