AI Briefing2026 Benchmark for 4B-Class Models
Key point
In an M3 Pro benchmark, Nemotron 3 Nano took first place among 4B-class models.
Details
Comparing the latest 3~4B-class models under identical conditions on an 18GB M3 Pro, NVIDIA Nemotron-3-Nano 4B took first place with an overall score of 85%.
- The test consisted of 39 tasks.
- The breakdown was 15 finance, 15 reasoning, and 9 code.
- Each combination was run 3 times and the median was used, with temp=0, seed=42, and max_tokens=1024 fixed.
Scores by model were as follows.
- nemotron-3-nano:4b: overall 85%, finance 100%, reasoning 80%, code 67%
- phi4-mini:3.8b: overall 77%, finance 80%, reasoning 60%, code 100%
- gemma4:e4b: overall 62%, finance 60%, reasoning 60%, code 67%
- granite4:3b: overall 54%, finance 60%, reasoning 20%, code 100%
- qwen3.5:4b: overall 15%, finance 20%, reasoning 20%, code 0%
Notably, Nemotron 3 Nano answered all 15 finance questions correctly, and finished its internal reasoning process with a </think> tag within the 1024-token limit. The author noted that this model outperformed larger models in the finance domain within the same size class.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Step 5 Preview Released: 600B Parameter MoE Model Highlights Agentic Tasks and Financial Analysis PerformanceQwen3.5-based Small Decision Model 'Kev' Released with TypeSafe API Compatibility and Performance Metrics