GPT-3.5 is history + MindStudio's new pricing & packaging

We finally made it: our new pricing packages are ready to go live next week on Monday 18th. Please look into your email to find the latest announcement and next steps!

This week, we might be witnessing the demise of a very popular model, aka RIP GPT-3.5. While it still makes sense for some use cases, and it’s obviously great to stick to what we know for existing workflows, the new Claude 3 Haiku surpasses it in every way other than function calling.

In other news, software engineers are freaked out by Devin, the first AI software engineer that can pass over 13% of the SWE-bench, outperforming all current models.

Keep reading to learn more!

Congratulations to our community members and creators that contributed to making this week special!

🗞️ Industry news

1. Anthropic Releases Claude 3 Haiku, the most versatile cost-efficient model.
   Anthropic released Claude 3 Haiku a couple of days ago, and we added it to the platform shortly after.
   
   Haiku is a revolutionary model, and probably the most important one for a wide variety of use cases in the Claude 3 family. Why? Mostly due to pricing.
   
   Historically, GPT-3.5 has been a great model for beginners to get their hands dirty building with AI. It was likely not going to stay the model of choice forever for enterprise apps, but it was a great starting point.
   

   

   
   With Haiku, that might be changing. Haiku is:
   
   - Up to 60% faster than GPT-3.5
   - More performant across the board
   - Significantly cheaper, offered at $0.33/1m tokens in input on our platform
   - 200k context window, vs GPT-3.5 at 16k
   
   The only reason I can think of where GPT-3.5 still outperforms Claude is the function calling API. Other than that, if your business workflows allow it, I highly suggest you take a look into Claude 3 Haiku and combinations of Haiku with higher intelligence models in the family like Sonnet and Opus.
   

2. Devin, the first AI-software engineer, is here and scares developers worldwide
   

   

   
   Software developers have been automating everyone else’s job for a while, but it always hits harder when we see an AI or robot do what we do. Devin did that with coding.
   
   Devin is NOT an LLM model, nor is likely as impressive as it looks in real-life scenarios, but it’s the first step towards agentive AI that can replace, not simply assist, coders.
   
   The company claims Devin was able to solve 13.86% of the issues on the SWE-bench (real open source coding challenges from Github) unassisted.
   
   Now, should developers actually be afraid? Opinions here vary. This is definitely not good enough to pose any real threat right now. It still scores lower than an entry software engineer, and can’t take responsibility for anything when things go wrong. But it most definitely will be in the future.
   
   That being said, all jobs eventually get replaced, and better ones are typically on the horizon. Less than 100 years ago, the vast majority of the population was working in farms barely surviving. Now, the largest sector of the economy is the service economy.
   
   Developers might be out of one job, but might find themselves in a better, more high level role soon. That’s the end goal of AI: make humans live better, more fulfilling lives, while increasing productivity and find new ways to discover.
   

3. OpenAI’s SORA isn’t releasing anytime soon
   OpenAI’s CTO Mira Murati confirmed SORA isn’t coming anytime soon. It will come “later this year”, but you shouldn’t expect the groundbreaking text-to-video model to be generally available in the next month or so like a few predicted.
   
   This is odd given OpenAI always released products shortly after their announcement, but it makes sense since AI videos could create massive problems worldwide, including in the 2024 US Election, with deep fakes and inappropriate usage of copyrighted material.

🔥 Product Updates

This week, we introduced several exciting features to further improve the platform:

• New Claude 3 Model: Claude 3 haiku is now available on all plans, and we highly suggest testing it out given it’s 60% faster than GPT-3.5!

• New pricing packages officially announced: the new pricing can be found in preview here. You will also receive a few emails reminding of the change. Please DO open them, there are several actions required for each account. New plans will change how and when you’re billed.

Together with the new pricing, we’re also releasing:

• New homepage: the new homepage will feature an AI chat interface and multiple tabs with your other open apps. It’s a much more streamlined experience that focuses on being there for you WHILE you work;

• All models available on all plans: after the switch to usage-based pricing, all models will be available on all plans. Free users will get access to premium models like Claude 3 Opus and GPT-4 Turbo (usage fees apply);

• API access: access to the API ships together with the pricing update on Monday 18th;

• Zapier integration: the Zapier integration will be available to all users shortly after, letting you build agentive AIs, apps that don’t require any interaction (e.g. schedulers, summarizers), and more.

💡 Tip of The Week

Do not obsess over token usage, but rather choose a model with a comfortable pricing level for you.

The difference between 1,000 and 10,000 tokens is very significant on Claude 3 Opus, but quite negligible on Claude 3 Haiku. Rather than using Opus everywhere trying to minimize the number of tokens used, try Haiku instead and use Opus only when required.

Every interaction with your application will use a unique number of tokens. It’s nearly impossible to predict with perfect accuracy. Hence, it’s much better to find a model stack and workflow you’re comfortable with regardless of the specific number.

We have a few FAQs on what’s a token and how to calculate token usage on our pricing page. A video is also coming out very soon explaining how to check usage in MindStudio.

🤝 Community Events

If you’re looking to meet our Product Manager and group up with community members, you can join our Discord Hangout on Friday, March 15th @ 3pm EST. Join from this link.

Our in-house MindStudio expert Giorgio Barilla will host a series of workshops over the next 6 weeks:

03/21 Customize AI-Powered Applications With Your Data

03/26 Build More Complex MindStudio Workflows

03/29 Connect AI-Powered Applications Across Your Tech Stack

Register to the ones that look interesting. We’ll upload the recordings for all of these in a future cohort.

🌯 That’s a wrap!

AI is moving faster than usual, and we’re happy to be part of this groundbreaking revolution. Stay tuned to learn more about what’s next and get tips & tricks for your MindStudio build.

You saw it here first,

Giorgio Barilla
MindStudio Developer & Project Manager @ MindStudio