Qwen2.5-Coder-7B-Instruct vs DeepSeek-Coder-V2-Lite-Instruct
Signed, community-submitted decode tok/s, prefill, and TTFT for Qwen2.5-Coder-7B-Instruct vs DeepSeek-Coder-V2-Lite-Instruct — every number links to the run it came from.
How to read these results
Selected submissions for RTX 5090 (32GB) report 255 tok/s on Qwen2.5-Coder-7B-Instruct and 309 tok/s on DeepSeek-Coder-V2-Lite-Instruct. The table selects each side's fastest submitted row independently, grouped by hardware label. Matching labels do not establish matching model artifacts, quantization, backend, workload or context. These peaks are reference results, not a controlled speed or answer-quality ranking. Inspect both source runs before choosing hardware.
Hardware with data on both models
| Hardware | Qwen2.5-Coder-7B-Instruct decode | DeepSeek-Coder-V2-Lite-Instruct decode | Δ | Source runs |
|---|---|---|---|---|
| RTX 5090 (32GB) | 255.3tok/s chat-short · llama.cpp · quantization unreported | 309.5tok/s chat-short · llama.cpp · quantization unreported | -54.2 | r_2b9o6y_49mi · r_0_gs1rgl2fl |
| M3 Ultra (60-core GPU) | 138.6tok/s chat-short · mlx · quantization unreported | 168.3tok/s chat-short · mlx · quantization unreported | -29.8 | r_uoehjq0nvc0 · r_l_v1-zq_qaz |
See also: Qwen2.5-Coder-7B-Instruct benchmarks · DeepSeek-Coder-V2-Lite-Instruct benchmarks · All hardware · All models · Methodology