Technology Videos

3 Big LLM Releases In 1 Week

The video highlights three major large‑language‑model (LLM) releases that landed within a single week, underscoring the accelerating pace of AI model innovation. First, DeepSeek unveiled two new variants—DeepSeek 3.2 and DeepSeek 3.2 Special—positioned as “reasoning‑first” models optimized for autonomous agents. The presenter emphasizes that these models were trained on a lean budget, contrasting sharply with the resource‑intensive training regimes of incumbent giants, and points to benchmark tables that show competitive performance despite the cost advantage. Next, the French research lab Mistral introduced its Mistral 3 family, which, according to the same benchmarks, matches the capabilities of DeepSeek 3.1 and the open‑source model Chem‑i K2. A key differentiator for Mistral is the Apache 2.0 licensing of all its models, granting developers unrestricted rights to run, fine‑tune, and deploy the models locally. This open‑source stance could lower entry barriers for enterprises seeking to embed LLMs without vendor lock‑in. The final segment teases an upcoming, unnamed model codenamed “Garlic,” reportedly in internal testing at a leading AI firm. Early internal evaluations suggest Garlic outperforms Gemini 3 and Opus 4.5 on coding and reasoning tasks. While the timeline is vague, the speaker speculates that the rapid development cycle could foreshadow a GPT 5.x release as early as next year, perhaps even within weeks, though he remains skeptical. Collectively, these announcements signal a shift toward more cost‑effective, open‑source, and specialized LLMs that could democratize access to advanced AI capabilities. Enterprises may soon have a broader palette of models to choose from—balancing performance, licensing flexibility, and operational expense—while the looming “Garlic” rumor hints at continued pressure on incumbents to accelerate their own roadmap.