Deepseek announced they are testing a new model.
A leaked test shows a massive new model crushing OpenAI's reading comprehension benchmark.
Deep Dive
A leaked benchmark test from Chinese AI company DeepSeek reveals they are testing a new model with a massive 256,000-token context window. The model reportedly achieved a score of 0.3595 on the OpenAI MRCR 8-pin reading comprehension test, suggesting it may outperform current benchmarks. The test data shows multiple perfect 1.0000 scores on specific prompts, indicating potentially state-of-the-art performance in long-context understanding and reasoning tasks.
Why It Matters
This leak signals a major new competitor capable of processing vastly more information than most current models, potentially shifting the AI landscape.