Hey there,

This week had a lot of AI news, so I’m trying something a little different. Instead of one longer deep dive, I’ll catch you up on the biggest AI news in just a few minutes. Let me know through the poll at the bottom if you like this new format. Don’t worry, we’ll be back to our usual deep dives next week. 

Claude Sonnet 5 is the new default

Anthropic released Sonnet 5 on June 30th and it has become the new default model for Claude. This means that Claude has received a significant upgrade for both free and paid users. Claude Sonnet 5 comes with increased agentic capabilities that allow it to utilize your terminal and web browsers to complete reasoning, coding, and knowledge work. This new model is comparable to Opus 4.8 in terms of performance but is a much cheaper option. Anthropic has also focused on reducing the model’s “misaligned behavior” and shows lower rates of hallucinations and sycophancy. Claude Sonnet 5 has already rolled out to everyone so head over to Claude and test it for yourself. 

GPT-5.6 went live as three tiers

OpenAI has made the GPT-5.6 family of models fully available on July 9th. This new model family consists of three new models, Sol, Luna, and Terra. Sol is the new powerful and expensive flagship model that's great for complex reasoning and coding tasks. Terra is the model for everyday tasks. Its performance level is similar to GPT-5.5 but much cheaper. Finally the Luna model is the cheap and fast lightweight model. This new family of models is meant to have an option for all potential use cases. The Luna and Terra models are available for all users while the Sol model is only accessible to paid users. GPT-5.6 is available through ChatGPT, Codex, and the API. 

Fable 5 is back after the government pulled it

Back in June we covered the release and eventual suspension of Anthropic’s Fable 5 and Mythos models. As of July 1st, Fable 5 access has been restored to all paid users. For anyone who needs a quick refresher, Fable 5 is a heavily constrained version of Mythos 5 and excels at complex reasoning and coding tasks. Its great performance metrics make it one of the best models currently available. Unfortunately Fable 5 will only be available to paid subscription users through July 12th, after which users with the annual or monthly subscription will lose access. Fable 5 will still be available through the API and usage-based credits. In the past Anthropic has said that they may reintroduce it into the subscription if rate limits allow.  

GPT-Live, a voice model you can interrupt 

OpenAI also released a new voice model this week called GPT-Live. This model is meant to feel more like an actual conversation by including the ability to listen and speak at the same time. They’ve also improved the flow by adding filler words and other common conversational habits. So now GPT-Live will say “hmmm” when thinking or stay quiet while you think about your response. The goal is to make conversations feel less rigid and more natural. GPT-Live is also able to push questions to frontier models like GPT-5.5 when needed. This ensures that the quality doesn’t degrade in an effort to get faster responses. 

My Thoughts

If there's one thing that’s jumped out at me this week, it's that we seem to be getting more and more models that are designed for specific tasks or use cases. It looks like the current AI trend is for companies to create several specialized models like GPT-Live but also multiple tiers of models like GPT-5.6 Sol, Luna, and Terra. It seems that these frontier labs want users to choose the best model for the job instead of always just going with the most powerful option. Personally, I’m someone who never really downgrades to a worse model unless I’ve hit my quota limits so I’m curious if users are manually doing this or if it's more just developers choosing which model they want to embed through the API.