Chip precision comparisons
46 head-to-head precision comparisons across DeepSeek V4 Pro 0813 1.6T, DeepSeek R1, Kimi K2.5/K2.6/K2.7-Code 1T, GLM 5/5.1, GLM 5.3 744B, MiniMax M3 428B, MiniMax M2.5/M2.7, Qwen 3.5 397B-A17B, and Llama 3.3 70B. See how FP4, FP8, BF16, INT4 and other quantization levels affect throughput, cost and interactivity on the same chip. Each page renders the inference chart and an interpolated comparison table.
DeepSeek V4 Pro 0813 1.6T
1 precision comparison with benchmark data on DeepSeek V4 Pro 0813 1.6T
DeepSeek R1
5 precision comparisons with benchmark data on DeepSeek R1
Kimi K2.5/K2.6/K2.7-Code 1T
3 precision comparisons with benchmark data on Kimi K2.5/K2.6/K2.7-Code 1T
GLM 5/5.1
5 precision comparisons with benchmark data on GLM 5/5.1
GLM 5.3 744B
3 precision comparisons with benchmark data on GLM 5.3 744B
MiniMax M3 428B
5 precision comparisons with benchmark data on MiniMax M3 428B
MiniMax M2.5/M2.7
5 precision comparisons with benchmark data on MiniMax M2.5/M2.7
Qwen 3.5 397B-A17B
13 precision comparisons with benchmark data on Qwen 3.5 397B-A17B
B200 — FP4 vs FP8
NVIDIA · Blackwell
B200 — FP4 vs BF16
NVIDIA · Blackwell
B200 — FP8 vs BF16
NVIDIA · Blackwell
B300 — FP4 vs FP8
NVIDIA · Blackwell
B300 — FP4 vs BF16
NVIDIA · Blackwell
B300 — FP8 vs BF16
NVIDIA · Blackwell
GB200 NVL72 — FP4 vs FP8
NVIDIA · Blackwell
GB300 NVL72 — FP4 vs FP8
NVIDIA · Blackwell
MI300X — FP8 vs BF16
AMD · CDNA 3
MI325X — FP8 vs BF16
AMD · CDNA 3
MI355X — FP4 vs FP8
AMD · CDNA 4
MI355X — FP4 vs BF16
AMD · CDNA 4
MI355X — FP8 vs BF16
AMD · CDNA 4
Llama 3.3 70B
2 precision comparisons with benchmark data on Llama 3.3 70B