DeepSeek R1 just got a HUGE Update! (o3 Level Model)

1. DeepSeek R1's update significantly enhances reasoning capabilities.

🥇95 00:26

The latest update improves DeepSeek R1's depth of reasoning and inference through enhanced computational resources and algorithmic optimizations.

• This update is considered substantial despite being labeled a minor upgrade.
• Performance benchmarks show improvements in mathematics, programming, and general logic.
• DeepSeek R1's performance is now comparable to leading models like OpenAI's 03.

2. DeepSeek R1 competes closely with leading AI models.

🥇92 00:50

DeepSeek R1's performance is now approaching that of top models, including Gemini 2.5 and OpenAI's 03.

• Benchmarks indicate that DeepSeek R1 has improved scores across various evaluations.
• It has shown significant advancements in coding skills, matching Gemini 2.5 Pro.
• The model is now recognized as a strong competitor in the AI landscape.

3. DeepSeek R1's architecture remains unchanged.

🥈88 02:35

Despite the performance leap, DeepSeek R1 retains its original architecture, indicating effective post-training refinements.

• The model has 671 billion parameters, with 37 billion active parameters.
• Post-training refinements have enhanced its reinforcement learning techniques.
• This approach has allowed for greater efficiency without altering the base model.

4. DeepSeek R1's token usage has increased for better performance.

🥈87 03:18

The new version of DeepSeek R1 utilizes more tokens for processing, enhancing its reasoning capabilities.

• It used 99 million tokens for evaluations, a 40% increase from the previous version.
• This increase allows for deeper thinking and more complex outputs.
• Gemini 2.5 Pro still uses more tokens, but DeepSeek R1's efficiency is notable.

5. The gap between open-source and closed-source models is narrowing.

🥇90 03:42

DeepSeek R1's advancements highlight the shrinking divide between open-source and proprietary AI models.

• DeepSeek R1 was initially a significant leap in open-source AI capabilities.
• The latest update positions it as a direct competitor to closed-source models.
• This trend reflects a growing parity in AI development between US and Chinese labs.