DeepSeek-V3-0324

DeepSeek-V3-0324 is an improved version of DeepSeek's super-powered language model, released on March 24, 2025. One of the key technical changes was a slight increase in the total number of parameters, from 671 billion in V3 to 685 billion in V3-0324.Compared to the previous version (DeepSeek-V3), the model shows significant improvements in logical reasoning, programming, and Chinese text generation. The model showed an impressive increase in results in key benchmarks:

• MMLU-Pro: from 75.9 to 81.2 (+5.3)
• GPQA: from 59.1 to 68.4 (+9.3)
• AIME: from 39.6 to 59.4 (+19.8)
• LiveCodeBench: from 39.2 to 49.2 (+10.0)

The developers emphasized that the model particularly stands out for its web application development capabilities - the generated code is not only executable but also has an improved visual design. DeepSeek-V3-0324 also demonstrates improved interactivity in multi-step dialogs, improves the accuracy of function calls and fixes bugs from the previous version, making it one of the most robust models in its class.

Announce Date: 24.03.2025
Parameters: 685B
Experts: 16
Activated: 37B
Context: 164K
Attention Type: Multi-head Latent Attention
VRAM requirements: 329.7 GB using 4 bits quantization
Developer: DeepSeek
Transformers Version: 4.46.3
Ollama Version: 0.5.5
License: MIT

Use our pre-built public endpoints to test inference and explore DeepSeek-V3-0324 capabilities.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.