AI digest: Enterprise tooling gets serious
Big moves in enterprise AI tooling as companies shift from experimentation to production deployment.
The enterprise AI game is heating up. This week brought proper tooling for production workloads and some serious strategic shifts.
OpenAI releases GPT-5.4 mini and nano for coding
OpenAI shipped two compact models built specifically for coding assistants and computer control. GPT-5.4 mini nearly matches the full model’s performance but costs up to 4x more than previous versions. This looks like OpenAI betting hard on enterprise use cases where performance matters more than cost.
Unsloth Studio tackles the VRAM problem
Unsloth AI released Unsloth Studio, a no-code interface for fine-tuning LLMs with 70% less VRAM usage. Finally, someone’s addressing the infrastructure pain of getting custom models running locally. This could democratise fine-tuning beyond teams with massive GPU budgets.
Mistral goes all-in on custom enterprise models
Mistral launched Mistral Forge at GTC, letting enterprises train completely custom models from scratch rather than just fine-tuning existing ones. Bold move considering most companies struggle to deploy basic chatbots properly. But if they can make it work, it’s a proper differentiator against OpenAI’s one-size-fits-all approach.
OpenAI expands government reach through AWS
OpenAI signed a partnership to sell AI systems to the US government through AWS for both classified and unclassified work. Smart distribution play, but given their recent strategic shift away from “side quests” to focus on enterprise customers, this feels like exactly the kind of high-value deal they need to justify their valuation.