Marina Lyubimova
⬤ DeepSeek reached a major milestone - its V3.2 model now ranks as the top open source system on Cortex-AGI, with a score of 38.2 percent and an overall sixth place. It stands beside the closed models of Google DeepMind, xAI, Anthropic besides OpenAI. The notable point is that DeepSeek beats multiple paid models while running at far lower cost.
⬤ The live board lists Gemini 3.0 Pro first at 45.6 percent, Grok 4.1-Thinking second at 42.6 percent or GPT-5.1-Thinking third at 41.1 percent. Cortex-AGI probes abstract reasoning through fresh logic puzzles that span ten difficulty levels - the test blocks memorization and targets real problem solving skill.
The benchmark strips away rote recall plus measures true generalization.
⬤ The outcome shows how the field is changing. Closed models keep the highest seats - yet DeepSeek sits above multiple commercial entries proving that open source code is closing the gap sooner than most predicted. The distance in high level reasoning keeps shrinking, a shift that may alter how firms pick and deploy models.
⬤ For the industry the practical message is that open source models now serve as valid options for demanding reasoning work once reserved for costly closed systems. As the performance gap narrows, price expectations and competitive pressure across the AI market will probably shift. The newest standings indicate that both sides are speeding up their progress pointing to a race that will likely tighten further in the coming months.
Marina Lyubimova
