Supported Models

This can subject to change

Jakey offers variety of models from Google, as well as other providers such as OpenAI, Mistral, Anthropic, through OpenRouter proxies and more. Understand these models' properties such as its hosting origin, model support, and use cases.

Currently Jakey supports models that is using serverless APIs, supports caching, and lower pricing. You can also provide your own API keys within the dev.env file or through environment variables according to the documentation of the SDK.

It uses LiteLLM for simplified model management and unified inferencing without the need of installing per-model SDKs, while Gemini uses the official SDK. OpenAI, Anthropic, Mistral models supports OpenRouter proxies when set and provided the provider API keys are not set.

To see the models live, use /model list command and set the default model with /model set command when using the Jakey @ experience

History and model usage

Because the APIs are different and doesn't use a unified solution like LangChain, conversation history are separated and branched while function calling name and prompt count remains shared.

For example, telling Gemini your name is John Smith would only make Gemini remember it as you use that model while GPT-4o doesn't remember what you've discussed with Gemini. Tool results, image inputs are also separated. Running /sweep command wipes all conversation across models.

Please see the table comparison below to see the breakthroughs

Available Jakey Models

The following models are available for use with Jakey, keep in mind that some models have its own limitations so choose what's best for your needs.

Gemini 1.5

Gemini 1.5 Pro and Flash offers the longest context window with balanced intelligence and speed with 1-2 million input tokens. This is the default model Jakey uses due to versatility. Here you can use Jakey with the following use cases:

• Analyze long and complex PDFs complete with images and retains its fidelity utilizing for the model to see, with long context
• Get specific excerpts of a certain video clip
• Transcribe or describe audio files with fraction of speed
• Chat tasks that require attention and memory retention
• Creative writing tasks

Available model versions:

• gemini-1.5-pro-002
• gemini-1.5-flash-002
• gemini-1.5-flash-8b
• gemini-1.5-pro-001
• gemini-1.5-flash-001

GPT-4o

GPT-4o offers the intelligence of GPT-4 and GPT-4 turbo (on par with Gemini 1.5 Flash May 2024 version for mini) with multimodal capabilities with a fraction of a price, in this version as opposed to the ChatGPT version of these models, it supports full 128K context window with multimodalities for both models, unlike the Free and Plus versions of ChatGPT where it is downscaled to 32K soft limit for input tokens. You can use this model with certain use cases such as:

• Get help with questions that requires broader knowledge
• Analyze images
• Draft certain types of data such as for content creation
• Summarize and get insights of a certain topic

Available model versions:

• gpt-4o - points to gpt-4o-2024-08-06
• gpt-4o-mini - gpt-4o-mini-2024-07-18

Currently GPT-4o only supports image inputs, with audio input and output coming soon.

Claude 3 models

Claude 3 models are latest iterations of the Anthropic model family. Ranging from intelligence and speed from Claude 3 Haiku to Claude 3.5 Sonnet, it surpasses most LLMs in terms of coding capabilities, creative human-like writing, and can tackle complex questions

Claude 3.5 Sonnet and 3.5 Sonnet (New) offers wide range of capabilities including:

• Complex coding such as writing and debugging high quality code surpasses all LLMs
• Creative writing
• Detailed responses
• Tasks that require multi-step planning and understanding
• Image analysis tasks

Available model versions:

• claude-3-5-sonnet-20241022 - Latest upgraded version of the 3.5 Sonnet, aliased with the -latest prefix.
• claude-3-5-sonnet-20240620 - The original version of 3.5 Sonnet.
• claude-3-5-haiku-20241022 - The smaller model of Claude 3.5 family surpassing Claude 3 Haiku
• claude-3-haiku - Points to latest 20240307 prefix.

XAI Grok

Grok 2 is a frontier language model that is on parity with GPT-4o, Claude 3.5 Sonnet, and LLaMA 405B. Try X.AI's Grok Models right within Jakey without buying a Twitter X premium plus badge endorsed subscription.

Keep in mind that the capabilities are the same as other frontier language models, as end users don't narrowly compare other flagship LLMs in terms of performance.

Available model versions:

• grok-beta - As of writing 11/5/2024 this is the only model available based on Grok 2. In this version, it doesn't have image capabilities, however vision models are onset next week and transition should be smooth as enabling the vision capabilities to Grok.

Mistral

Jakey offers mistral models and its superior for language and coding specific tasks as well of its reasoning capabilities but it remains on-par with other commercial models like GPT-4o and outclassed by Claude 3.5 sonnet

• Gather insights of specific information
• Assist with natural language tasks that involves multilingual capabilities
• Get help with coding and draft code

Available model versions:

• mistral-large-2407
• open-mixtral-8x7b
• codestral-latest

Table comparison

Models	Context size	Supports file attachments	Supports tools	Provider	Endpoint
- Gemini 1.5 Pro - Gemini 1.5 Flash - Gemini 1.5 Pro (001) - Gemini 1.5 Flash (001) - Gemini 1.5 Flash 8B	Pro: 2,000,000 Flash: 1,000,000	Yes - Images - Audio - Video - PDFs with optionally non-textual data - Source files (py, js, html, md, txt)	Yes	Google	Gemini API
- GPT-4o - GPT-4o Mini	128,000	Images only	No	OpenAI	- OpenAI API Platform - OpenRouter - Custom OpenAI Endpoint
- Claude 3.5 Sonnet v2 - Claude 3.5 Sonnet v1 - Claude 3.5 Haiku - Claude 3 Haiku	200,000	Images only (except Claude 3.5 Haiku)	No	Anthropic	- Anthropic API - OpenRouter
Grok (grok-beta)	128,000	No	No	XAI	XAI Console
Mistral Large v2	128,000	No	No	Mistral	- La Platforme - OpenRouter
Mixtral 8x7B	32,000	No	No	Mistral	- La Platforme - OpenRouter
Codestral	32,000	No	No	Mistral	La Platforme