June was a bit lighter on AI news for the first time this year, but a few things are still worth a mention.
So why was this month quieter? In earlier months we watched several big releases land at once, and now it has gone rather quiet. Companies tend to be consistent about when they release models, and right now they are back to improving them. The start of summer may have played a part. It may also be the first sign that scaling models is not everything.
Gemini 2.5 Pro and Flash
Both models are now stable and production-ready.
- You can use them in applications that serve end users, in other words, applications in production.
- Google is also releasing Flash 2.5 Lite, a faster version of Flash. Because of pricing changes, though, it does not always make sense to use it.
Useful links: Deepmind Google
ChatGPT adds connectors
ChatGPT now supports Hubspot, Salesforce, Teams, and more.
- OpenAI plans to keep expanding the MCP server network and let users work with application data inside ChatGPT. This may have security implications.
Useful links: ChatGPT for Business Updates
DeepSearch via API
DeepSearch is now usable in applications built by third-party developers too.
- It has a lot of potential, and what it amounts to depends on what developers build on top of it.
Cursor ships background agents
Cursor lets you run some programming tasks straight from Slack.
- It also brings a number of security risks.
- It points to the need for usage rules at the company level.
Useful links: Docs Cursor
Weaknesses of LLM applications
June also brought attention to the weaknesses of applications built on LLMs:
- Attacks on agents through "poisoned" prompts are starting to appear.
- Defending against them is extremely hard, and a fake MCP server is not always to blame.
- The attack assumes the LLM can execute commands on its own and the attacker can reach it to send the prompt.
Inception Labs releases the Mercury model
The Mercury model is built on a diffusion architecture.
- It matches the performance of GPT-4.1 Nano but is 7x faster, at roughly the same price.
- Models of this kind could be the future.
Useful links: LinkedIn Post Artificial Analysis
June showed that interesting things happen even in a quieter month. Gemini models are ready for production, ChatGPT is expanding its connectors, and new challenges are emerging around the security of LLM-based applications. Inception Labs' Mercury hints that diffusion architectures could be the future of fast AI models.
Want to stay one step ahead?
Don't miss our best insights. No spam, just practical analyses, invitations to exclusive events, and podcast summaries delivered straight to your inbox.