Competition Among Latest LLMs Including DeepSeek, Mistral, Gemini and Advances in Text-to-Video Models
Acceleration of AI Agent and Workflow Integration Including Google Workspace Studio and Anthropic Interviewer, and Model Refinement Strategies

Latest Model Releases and Performance Competition

DeepSeek released its V3.2 model claiming performance matching GPT-5, with the V3.2-Speciale variant featuring higher computational power competing with Gemini-3.0-Pro and winning gold medals at IMO, IOI, and ICPC 2025. Mistral also released a new model family, Mistral 3, including 3 dense models (14B, 8B, 3B) and Mistral Large 3, a sparse MoE model with 41B active parameters. All Mistral 3 models are available open source under Apache 2.0 license.

In text-to-video generation, Runway Gen-4.5 achieved the top score on Artificial Analysis benchmarks, surpassing Veo 3 and Sora. The model emphasizes physical accuracy including realistic momentum, fluid dynamics movement, and material consistency, while acknowledging persistent challenges like object permanence. STARFlow and STARFlow-V also introduced transformer autoregressive flow architecture for high-quality image and video generation, combining the expressiveness of autoregressive models with the efficiency of normalizing flows.

Google''s Gemini 3 Deep Think is now available in the Gemini app, offering parallel reasoning to simultaneously explore multiple hypotheses. The model is based on the Gemini 2.5 Deep Think variant that won a gold medal at the International Mathematical Olympiad. OpenAI also presented the optimal approach for GPT-5.1-Codex-Max, featuring enhanced compression capabilities for faster token efficiency, long-term autonomy, and extended reasoning.