Best local AI model for RTX 5090 32GB.

Qwen3.6-35B-A3B (higher quant)

Use Q5-ish / FP4 where supported, 32k-64k practical. This is the highest-scoring current open-weight model that fits this card cleanly, selected by benchmark then fit then freshness, not by parameter count.

Same Qwen3.6-35B-A3B score profile (MMLU-Pro 85.6/85.0, GPQA Diamond 84.9/84.8); NVFP4 loses little vs BF16, which matters for Blackwell-era deployment.

NVFP4 numbers show low degradation vs BF16; on 32GB the lever is quality and context, not parameter count.

Qwen3.6-35B-A3B at FP4 for context headroom

Qwen3-32B dense at higher quant

70B only as a compromised low-quant test