DeepSeek cloud models

31 models tracked via Artificial Analysis. Compare cloud performance, then find local GGUF versions in the GraySoft model catalog.

Model	Intelligence	Speed (tok/s)
DeepSeek V4 Pro (Reasoning, Max Effort)	44.3	71.807
DeepSeek V4 Pro (Reasoning, High Effort)	43.1	75.202
DeepSeek V4 Flash (Reasoning, Max Effort)	40.3	125.4
DeepSeek V4 Flash (Reasoning, High Effort)	37.5	0
DeepSeek V3.2 (Reasoning)	32	0
DeepSeek V4 Pro (Non-reasoning)	31.2	70.522
DeepSeek V3.1 Terminus (Reasoning)	30.4	0
DeepSeek V4 Flash (Non-reasoning)	28.7	130.333
DeepSeek V3.2 Exp (Reasoning)	25.4	0
DeepSeek V3.2 (Non-reasoning)	24.7	0
DeepSeek V3.2 Speciale	22.2	0
DeepSeek V3.1 Terminus (Non-reasoning)	21.4	0
DeepSeek V3.2 Exp (Non-reasoning)	21.3	0
DeepSeek V3.1 (Non-reasoning)	21	0
DeepSeek V3.1 (Reasoning)	20.7	0
DeepSeek R1 0528 (May '25)	20.1	0
DeepSeek R1 (Jan '25)	18.5	0
DeepSeek V3 0324	15.4	0
DeepSeek V3 (Dec '24)	14.2	0
DeepSeek R1 Distill Qwen 32B	11	0
DeepSeek R1 0528 Qwen3 8B	10.4	0
DeepSeek R1 Distill Llama 70B	9.9	0
DeepSeek R1 Distill Qwen 14B	9.8	0
DeepSeek-V2.5 (Dec '24)	6.8	0
DeepSeek-V2.5	6.6	0
DeepSeek R1 Distill Llama 8B	6.4	0
DeepSeek-Coder-V2	5.1	0
DeepSeek R1 Distill Qwen 1.5B	3.7	0
DeepSeek-V2-Chat	3.6	0
DeepSeek Coder V2 Lite Instruct	3.1	0
DeepSeek LLM 67B Chat (V1)	3	0

Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.