# AI News Roundup: OpenAI’s Upcoming Models, Open Source Developments, and Industry Shifts
## OpenAI’s Upcoming Releases
OpenAI is on the verge of releasing several new models. Sam Altman recently hinted at “good stuff coming this week,” with leaked API assets suggesting we might see GPT-4.1, GPT-4.1 Nano, GPT-4.1 Mini, or O4 Mini as soon as tomorrow.
In a recent podcast, Altman shared insights about their development process:
– Work on GPT-4.5 began two years ago
– GPT-5 is turning out “much better than originally thought” but integration is proving challenging
– They’re facing “unprecedented demand,” particularly for GPT-4o image generation
– GPU shortages are a significant constraint
– The O3 model has been improved “in many ways” from what was previewed
## Memory: OpenAI’s New Personalization Strategy
OpenAI released “essentially infinite memory” for ChatGPT, allowing it to reference all past conversations to deliver more personalized responses. This suggests a strategic pivot:
– As raw intelligence becomes commoditized across the industry, OpenAI is differentiating through personalization
– This aligns with research showing therapy and companionship are now the #1 use cases for AI
– Memory is available to Plus and Pro users, with full user control over what information is retained
## Open Source Advancements
The open source community continues to make impressive strides:
1. **DeepCoder 14B**: Together AI and Aentica released a fully open source coding model matching O3 mini low performance but with much higher efficiency.
2. **Cogito V1**: DeepCog launched a new open source model based on Llama 3 70B that outperforms its base model across benchmarks, featuring “thinking tokens” and enhanced multilingual coding capabilities.
## Midjourney V7 Alpha
Midjourney quietly released V7 alpha with significant improvements:
– Smarter handling of text and image prompts
– Noticeably higher image quality
– New “draft mode” that renders images 10x faster at half the cost