DeepSeek V3-0324 has set a new benchmark in open-source AI, becoming the highest-scoring non-reasoning model on the Artificial Analysis Intelligence Index — a breakthrough achievement that signals a growing shift in AI innovation.
The new model climbed seven points in the latest benchmark update, surpassing powerful proprietary contenders such as Google’s Gemini 2.0 Pro, Anthropic’s Claude 3.7 Sonnet, and Meta’s Llama 3.3 70B. This marks the first time an open weights model has claimed the top spot in the non-reasoning category — a major milestone for the open-source AI movement.
Unlike complex reasoning models, non-reasoning models are designed for real-time performance. They generate fast, responsive outputs without requiring deliberative “thinking” stages, making them ideal for latency-sensitive use cases such as:
- Customer service automation
- Real-time chatbots
- Instant translation tools
- Voice assistants
With its latest update, the DeepSeek V3-0324 open-source AI model now leads this segment — outperforming not just its open-source peers but also surpassing leading proprietary non-reasoning models in speed and accuracy.
“This is the first time an open weights model leads in non-reasoning performance,” Artificial Analysis confirmed in its latest report. “It’s a huge step forward for open-source AI.”
While V3-0324 represents a major leap in open-source capability, running the model requires serious computing power:
- 128k context window (API-capped at 64k)
- 671 billion total parameters needing over 700GB GPU memory (FP8 precision)
- 37 billion active parameters
- Text-only functionality (no multimodal support)
- Released under the MIT License
“Still not something you can run at home,” Artificial Analysis joked — highlighting the model’s enterprise-level infrastructure requirements.
Despite this, the model’s open-source status makes it an attractive option for developers seeking adaptability and transparency in building high-performance AI systems.
While DeepSeek’s reasoning model R1 still dominates in tasks requiring complex problem-solving, V3-0324’s latest performance demonstrates that open-source models are rapidly closing the gap.
Just three months ago, DeepSeek V3 fell slightly behind Anthropic and Google’s closed models. Now, with the release of V3-0324, the model not only leads the open-source pack but also overtakes proprietary non-reasoning competitors.
“This release is arguably even more impressive than R1,” Artificial Analysis noted, praising the technical leap.
For enterprises and AI developers, this milestone is significant. DeepSeek V3-0324 offers an MIT-licensed, powerful, and highly adaptable AI tool — though its computational demands may limit immediate accessibility for smaller teams.
With DeepSeek R2 reportedly on the horizon, the open-source community is watching closely for the next leap in performance. The company’s rapid advancements suggest that open-source AI is no longer just playing catch-up — it’s now driving the frontier of model development.
“DeepSeek is pushing the boundaries for non-reasoning open weights models,” Artificial Analysis declared. “This could reshape the landscape for developers looking for high-performance, real-time AI without being locked into proprietary ecosystems.”
