RTX 4070 Ti Super vs RTX 5070 Ti for local LLMs

Prices are US launch prices (MSRP) for comparison; street prices change daily. Links go to Amazon, where we earn a commission at no cost to you.

Every model, side by side

Best quant that fits and estimated speed at 8K context.

Model	RTX 4070 Ti Super	RTX 5070 Ti
Qwen3 0.6B	FP16 · ~269 tok/s	FP16 · ~358 tok/s
Gemma 3 1B	FP16 · ~202 tok/s	FP16 · ~269 tok/s
Llama 3.2 1B	FP16 · ~163 tok/s	FP16 · ~217 tok/s
Qwen3 1.7B	FP16 · ~99 tok/s	FP16 · ~132 tok/s
Llama 3.2 3B	FP16 · ~63 tok/s	FP16 · ~84 tok/s
Phi-3.5 Mini 3.8B	FP16 · ~53 tok/s	FP16 · ~70 tok/s
Phi-4 Mini 3.8B	FP16 · ~53 tok/s	FP16 · ~70 tok/s
Qwen3 4B	FP16 · ~50 tok/s	FP16 · ~67 tok/s
Gemma 3 4B	FP16 · ~47 tok/s	FP16 · ~63 tok/s
Mistral 7B v0.3	Q8_0 · ~52 tok/s	Q8_0 · ~70 tok/s
Qwen2.5 7B	Q8_0 · ~50 tok/s	Q8_0 · ~66 tok/s
Qwen2.5 Coder 7B	Q8_0 · ~50 tok/s	Q8_0 · ~66 tok/s
DeepSeek R1 Distill Qwen 7B	Q8_0 · ~50 tok/s	Q8_0 · ~66 tok/s
Llama 3.1 8B	Q8_0 · ~47 tok/s	Q8_0 · ~63 tok/s
Granite 3.3 8B	Q8_0 · ~46 tok/s	Q8_0 · ~62 tok/s
Qwen3 8B	Q8_0 · ~46 tok/s	Q8_0 · ~62 tok/s
DeepSeek R1 0528 Qwen3 8B	Q8_0 · ~46 tok/s	Q8_0 · ~62 tok/s
Gemma 2 9B	Q8_0 · ~41 tok/s	Q8_0 · ~55 tok/s
Gemma 4 12B	Q8_0 · ~32 tok/s	Q8_0 · ~42 tok/s
Mistral Nemo 12B	Q8_0 · ~31 tok/s	Q8_0 · ~41 tok/s
Gemma 3 12B	Q8_0 · ~31 tok/s	Q8_0 · ~41 tok/s
Qwen2.5 14B	Q6_K · ~33 tok/s	Q6_K · ~45 tok/s
Phi-4 14B	Q6_K · ~33 tok/s	Q6_K · ~45 tok/s
Phi-4 Reasoning 14B	Q6_K · ~33 tok/s	Q6_K · ~45 tok/s
Qwen3 14B	Q6_K · ~33 tok/s	Q6_K · ~44 tok/s
Qwen2.5 Coder 14B	Q6_K · ~33 tok/s	Q6_K · ~44 tok/s
DeepSeek R1 Distill Qwen 14B	Q6_K · ~33 tok/s	Q6_K · ~44 tok/s
gpt-oss 20B	Q5_K_M · ~157 tok/s	Q5_K_M · ~210 tok/s
Mistral Small 24B	Q3_K_M · ~35 tok/s	Q3_K_M · ~47 tok/s
Devstral Small 24B	Q3_K_M · ~35 tok/s	Q3_K_M · ~47 tok/s
Magistral Small 24B	Q3_K_M · ~35 tok/s	Q3_K_M · ~47 tok/s
Mistral Small 3.2 24B	Q3_K_M · ~34 tok/s	Q3_K_M · ~46 tok/s
Gemma 4 26B-A4B	Q3_K_M · ~217 tok/s	Q3_K_M · ~289 tok/s
Gemma 2 27B	Q2_K · ~40 tok/s	Q2_K · ~53 tok/s
Gemma 3 27B	Q3_K_M · ~30 tok/s	Q3_K_M · ~40 tok/s
Qwen3.8 27B	Q3_K_M · ~30 tok/s	Q3_K_M · ~40 tok/s
Qwen3.6 27B	Q3_K_M · ~30 tok/s	Q3_K_M · ~40 tok/s
Qwen3 30B-A3B	Q2_K · ~326 tok/s	Q2_K · ~434 tok/s
Qwen3 Coder 30B-A3B	Q2_K · ~326 tok/s	Q2_K · ~434 tok/s
GLM-4.7 Flash 30B-A3B	Q2_K · ~358 tok/s	Q2_K · ~478 tok/s
Gemma 4 31B	Q2_K · ~34 tok/s	Q2_K · ~46 tok/s
Nemotron 3 Nano 30B-A3B	Q2_K · ~307 tok/s	Q2_K · ~410 tok/s
Qwen2.5 32B	Q2_K · ~33 tok/s	Q2_K · ~44 tok/s
DeepSeek R1 Distill Qwen 32B	Q2_K · ~33 tok/s	Q2_K · ~44 tok/s
Qwen3 32B	Q2_K · ~33 tok/s	Q2_K · ~44 tok/s
QwQ 32B	Q2_K · ~33 tok/s	Q2_K · ~44 tok/s
Qwen2.5 Coder 32B	Q2_K · ~33 tok/s	Q2_K · ~44 tok/s
Qwen3.6 35B-A3B	Q2_K · ~358 tok/s	Q2_K · ~478 tok/s
Mixtral 8x7B	Too big	Too big
Llama 3.1 70B	Too big	Too big
Llama 3.3 70B	Too big	Too big
DeepSeek R1 Distill Llama 70B	Too big	Too big
Qwen2.5 72B	Too big	Too big
Qwen3 Coder Next 80B-A3B	Too big	Too big
GLM-4.5-Air 106B-A12B	Too big	Too big
Llama 4 Scout 17B-16E	Too big	Too big
gpt-oss 120B	Too big	Too big
Qwen3.8 Flash Next	Too big	Too big
Qwen3 235B-A22B	Too big	Too big
DeepSeek V4 Flash 284B-A13B	Too big	Too big
GLM-5.3 Flash 320B-A18B	Too big	Too big
MiniMax M3 428B-A23B	Too big	Too big
GLM-5 744B-A40B	Too big	Too big
GLM-5.3	Too big	Too big
DeepSeek V4.1 Flash	Too big	Too big
Mistral Large 4 1T-A52B	Too big	Too big
DeepSeek V4 Pro 1.6T-A49B	Too big	Too big
Qwen3.8 2.4T-A95B	Too big	Too big
Kimi K3.1 Pro 2.8T-A104B	Too big	Too big

More comparisons

• RTX 4090 vs RTX 5090
• RTX 3090 vs RTX 4090
• RTX 3090 vs RTX 5090
• RTX 5080 vs RTX 4090
• Radeon RX 7900 XTX vs RTX 4090
• Radeon RX 7900 XTX vs RTX 3090
• RTX 4060 Ti 16GB vs RTX 5060 Ti 16GB
• RTX 5070 vs RTX 5060 Ti 16GB
• RTX 3060 12GB vs RTX 4060
• Apple M4 Max (128GB) vs RTX 5090
• Apple M4 Pro (48GB) vs RTX 4090