Reading the model card for Gemini 3, you'll find that Gemini 3 is clearly not a fine-tuning of Gemini 2.5; it is a newly trained sparse MoE. In other words, based on the already outstanding RL post-training and parallel thinking of Gemini 2.5, the brand new backbone makes Gemini 3 exceptional. Summarizing the work of Gemini over the past six months: 1. Outstanding RL post-training 2. Parallel thinking 3. Brand new backbone 4. One beneficial public benchmark for the industry after another, such as IMO-Bench Now, let's compare what OpenAI has done in the past six months: 1. Nonexistent GPT-5 2. Adaptive thinking, focusing on saving money for enterprises while treating users like fools 3. The forgotten agent mode 4. Proactive assistant, driven by a surge of advertising 5. Credit expiration, keeping a close eye on users' wallets 6. Sam's endless hype What defines a great company, and what defines a sneaky company?