How to Connect to Anthropic’s Claude API and Groq’s API

by Matt Zimmerman

Updated May 10, 2024

The AI in ZimmWriter isn’t limited to OpenAI anymore. You can now utilize Anthropic’s ‘Claude’ models as well as certain models from Groq.

However, it’s important to note that ZimmWriter was originally built on OpenAI’s models. Each model has its own quirks and issues. Although I’ve tried to patch around things that produce unexpected outputs, you might still encounter glitches in the blog posts when using these models. Therefore, make sure to review your articles carefully.

Anthropic’s API

Anthropic, like OpenAI, is a company that provides AI models which ZimmWriter can use to write blog posts.

How to Get an Anthropic API Key

First, head over to Anthropic’s website and sign up to receive an API key.

Upon creating an account, you’ll automatically enter the free tier. Like OpenAI, Anthropic enforces usage and rate limits. To quickly advance to tier 2, which is ideal for running a single ZimmWriter license, simply add $50 to your account. Although the usage and rate limit page mentions a 7-day waiting period, I was immediately upgraded, so you might have similar luck.

If you need to run multiple licenses of ZimmWriter on different computers, consider moving up to higher tiers to accommodate extensive use of Anthropic’s Claude models.

On the Anthropic website, you’ll also find an option to generate an API key. Once you have it, go to the ZimmWriter options menu and click ‘Setup Text API (Non-OpenAI)’. Enter your key there, and you’re all set.

Anthropic’s Claude AI Models

At the time of writing this article, Anthropic offers three models that ZimmWriter can currently utilize.

  • Claude 3 Opus is the most expensive model, comparable in price and quality to GPT-4.
  • Claude 3 Sonnet is the mid-range model, similar to GPT-4 Turbo.
  • Claude 3 Haiku is the least expensive model, comparable to GPT-3.5.

You can find specific pricing for these models on Anthropic’s pricing page. Each model’s pricing is listed for both ‘input’ tokens and ‘output’ tokens.

What are input and output tokens?

When you interact with the AI, we need to input your request, consuming input tokens, and then the AI responds, consuming output tokens.

However, the prices on the page are listed in millions of tokens—for example, Claude 3 Sonnet costs $15 per million output tokens. But how does this translate to the cost of writing articles?

It’s challenging to determine precisely because the cost varies significantly depending on the settings you choose for writing your articles. Generally, a rough estimate is that one million tokens for input and output combined could produce about 50-75 bulk writer articles, each with 10 H2 headers, FAQs, no H3 headers, no SERP scraping, and no custom outlines.

Groq’s API

Groq’s another company whose AI models you can use. The main difference is that Groq’s models are Facebook’s open-source models. Currently, the only model powerful enough to run ZimmWriter’s prompts is the LLaMA3 70b model.

You can register for an account and get your API key on Groq’s API page.

Now the really amazing thing (at least at the time of writing this) is that you can use Groq’s API for free right now! Soon it’ll switch to a paid mode, but for now, it’s free.

But… (there’s always a but).

The free plan is rate limited, so it’s going to go slower than OpenAI or Anthropic’s models. But when it goes paid, it should be the fastest generator on the market.

Groq’s AI Models

The only model capable of running ZimmWriter prompts is LLaMA3 70b.

People rank the quality of LLaMA3 70b higher than GPT4 Turbo but slightly below GPT4.

Now, you might be wondering why you can’t simply run this open-source model developed by Facebook on your personal computer for free.

That’s an excellent question.

I’ve tested the LLaMA3 8b model, which has 8 billion parameters and should technically run on most computers, but it fails to handle the prompts, producing unusable results. Therefore, we must rely on the 70 billion parameter model.

However, the LLaMA3 70b model requires 34GB of GPU RAM, exceeding the 24GB maximum found in consumer-grade graphics cards. I’m uncertain about using SLI configurations, but that’s likely beyond what most users possess.

Running it on PC RAM isn’t viable either due to memory bandwidth limitations.

You might think that M-series Mac computers, with their superior memory bandwidth, could be a solution. Why not run it there using ZimmWriter on Parallels or Wine?

Another great question.

Unfortunately, that approach also falls short. Even my M2 Max with 96GB RAM struggles with the LLaMA3 70b model, making it impractically slow. Plus, how many of you have M-series Macs with over 64GB of RAM?

For now, Groq remains an excellent alternative.

Even when Groq starts charging, they’ve indicated their pricing will be competitive with or less than OpenAI’s GPT 3.5.

Switch to GPT 3.5 on Server Errors

In the ZimmWriter options menu, where you configure API keys, you’ll find an interesting option titled ‘Switch to GPT 3.5 Turbo on Server Errors for a Particular Job’.

If you’re using Claude and you run out of credit, or if Claude’s servers go down mid-job and they’re not expected to be up for a few days, what do you do? Especially if you’re in the middle of generating 1,000 blog posts?

By default, ZimmWriter will show a pop-up message giving you three choices:

  1. Retry (ZimmWriter will automatically retry every 10 minutes, or you can press the retry button to manually retry) up to 99 times.
  2. Switch to OpenAI’s GPT 3.5 for the remainder of your job (e.g., writing 1,000 blog posts).
  3. Exit ZimmWriter.

However, you might not want ZimmWriter to auto-retry every ten minutes. Perhaps you’d prefer it to immediately start using GPT 3.5 Turbo and finish the job. If that’s the case, simply check this box. When checked, ZimmWriter will automatically switch to OpenAI’s models for the remainder of the job, whenever Anthropic or Groq have issues.

ranking tactics logo

The information provided on this website is provided for entertainment purposes only. I make no representations or warranties of any kind, expressed or implied, about the completeness, accuracy, adequacy, legality, usefulness, reliability, suitability, or availability of the information, or about anything else. Any reliance you place on the information is therefore strictly at your own risk. Read more in my terms of use and privacy policy. You can also contact me with questions.

Owned and Operated by Revindir LLC
Copyright 2024