GPT-4.5, Sonnet 3.7, Google's Code Assistant, Amazon brings AI to Alexa - 2nd March 2025

Phew, what a busy week it's been in the world of AI.

OpenAI releases GPT-4.5

We also know that GPT-5 is expected in the coming few months according to Sam Altman.

Although GPT-4.5 is currently only available to USD $200/month ChatGPT subscribers (and via API), OpenAI are claiming their new GPT-4.5 model:

• has lower hallucination rate as the model is larger and inherently knows more
• Improved pattern recognition and adherence to complex instructions
• Increased creativity and emotional intelligence
  • Perhaps the next evolution in AI mental health support
• Currently very expensive to access via API ($USD):
  • $75.00/1M input tokens
    • (vs. $2.50/1M input tokens for GPT-4o)
  • $150.00/1M output tokens
    • (vs. $10.00 / 1M output tokens for GPT-4o)

My 2cents on OpenAI's GPT-4.5:

Always great to see a new release, especially one that seems to have a higher emotional quotient and more knowledge.

However it does feel that OpenAI are starting to clutch at straws with the competition from other frontier labs really heating up.

It's also expected that GPT-5 isn't going to be a huge jump, but rather act as a "model router" of sorts. Meaning instead of the user picking the best model for the job they simply give the AI their request or question and the AI picks the best models and tools to achieve that request.

OpenAI keeping functionality locked behind a $200 / month pay wall is also an interesting play when competitions are releasing similar functionality for lower prices.

Anthropic Released Claude 3.7 Sonnet and Claude Code

Claude 3.7 Sonnet is a huge leap forward for Anthropic. It's got strong reasoning capabilities that are also transparent to the user, showing what it's thinking. Claude 3.7 Sonnet is also available on all paid plans which is great to see.

• Improved accuracy in mathematical proofs by 18%
• Improved coding task accuracy of 23%

What I think is really exciting however is the release of Claude Code. Sonnet has already been the preferred model for many developers so it makes sense that the Anthropic team are really doubling down on this.

Claude Code operates directly from the CLI and can be used to help refactor, document and debug code.

My 2cents on Claude 3.7 Sonnet and Claude Code:

It's great to see new releases that directly improve developers' efficiency and accuracy. I personally use ChatGPT as my go-to browser based AI chat but I've been using Claude's models in my day to day coding inside Cursor.

It's exciting and a bit scary to see how quickly developer AI tools and AI coding assistants are evolving. This is an industry that is in for some rapid change.

Google's Gemini Released a Coding Assistant

Google has released their "Cursor/Copilot Killer" - kind of:

• Generous free tier with 180k code completions per month
  • Microsoft's Copilot gives just 2k on their free tier for comparison
  • It really seems like Google is trying to use a lolly scramble to bribe devs while they play AI catchup...
• Integrates with Visual Studio Code and the JetBrains's family of IDEs
• Automatic test generation - probably my favourite use case for AI coding assistants
• Powered by Gemini 2.0 behind the scenes, probably their faster flash model for inline completion but using the slower, smarter model for chat

My 2cents on Gemini's Coding Assistant

More competition is always great for the end user especially in the dev tools space.

Personally I don't plan to switch from Cursor anytime soon (using Claude's model) but I do like seeing Google offering a generous free tier.

This could be especially useful for students and lower income folks trying to learn coding who might currently be priced out of the $20+ per month fee Copilot and Cursor charge.

Amazon has released Alexa+ with AI

After numerous delays and technical challenges, Amazon has finally released their Alexa+ (Alexa but with AI).

• Prices at USD $19.99/month or included for Amazon Prime subscribers
• Alexa+ uses LLMs to make for a smarter Alexa experience
• Operates on AWS Bedrock, allowing for multiple models to be used under the hood
• Uses a vector database to store user preferences such as dietary restrictions and general preferences
• Proactive task automation which sounds both epic and scary
  • basically it can preempt what you might want and go ahead and book or buy things for you

My 2cents on Alexa+:

I think it's great that we're seeing a major player in the voice assistant game leverage AI to make a far superior product.

This is an area of the market I still think the tech is slower to catch on in but actually is one of the most impactful areas of our lives for AI.

Many of us are used to using Siri, Alexa, Google assistant, etc whether it be on your devices or in our homes so adding AI to existing behaviour is a great way to encourage uptake.