DeepSeek V4 Pro 1.6T
8K/1K
B200
vLLM · FP4
MI355X
B300
vLLM · FP4
GB200 NVL72
Dynamo vLLM · FP4
GB300 NVL72
—
no exact @30 result
Inference Cost per Million Tokens
Hyperscaler cost ↓ Lower is better Source: InferenceX & SemiAnalysis Market July 2026 AI Cloud TCO Model
| Model · Scenario | B200 | MI355X | B300 | GB200 NVL72 | GB300 NVL72 |
|---|---|---|---|---|---|
DeepSeek V4 Pro 1.6T8K/1K | vLLM · FP4 | vLLM · FP4 | Dynamo vLLM · FP4 | — no exact @30 result | |
DeepSeek V4 Pro 1.6TLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | vLLM · FP4 | vLLM · FP4 | vLLM · FP4 | — no data for this scenario | — no exact @30 result |
Kimi K3 2.8TLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | Dynamo vLLM · FP4 · STP | — no data for this scenario | vLLM · FP4 | Dynamo vLLM · FP4 | — no data for this scenario |
MiniMax M3 428B8K/1K | ATOM¹ · FP4 | Dynamo vLLM · FP8 · STP | — no exact @30 result | ||
MiniMax M3 428BLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | — no exact @30 result | — no data for this scenario | — no exact @30 result | — no data for this scenario | — no data for this scenario |
GLM5.2Long Context Multi-Turn Realistic Agentic Scenario (AgentX) | — no exact @30 result | — no data for this scenario | — no exact @30 result | — no data for this scenario | — no data for this scenario |
Qwen3.5 397B8K/1K | TRTLLM · FP4 | SGLang · FP8 | — no exact @30 result | — no exact @30 result | |
Qwen3.5 397BLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | — no exact @30 result | — no data for this scenario | — no exact @30 result | — no data for this scenario | Dynamo SGLang · FP4 |
DeepSeek V4 Pro 1.6T
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
Kimi K3 2.8T
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
MiniMax M3 428B
8K/1K
B200MI355XATOM¹ · FP4B300GB200 NVL72Dynamo vLLM · FP8STPGB300 NVL72—no exact @30 resultMiniMax M3 428B
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
B200—no exact @30 resultMI355X—no data for this scenarioB300—no exact @30 resultGB200 NVL72—no data for this scenarioGB300 NVL72—no data for this scenarioGLM5.2
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
B200—no exact @30 resultMI355X—no data for this scenarioB300—no exact @30 resultGB200 NVL72—no data for this scenarioGB300 NVL72—no data for this scenarioQwen3.5 397B
8K/1K
B200TRTLLM · FP4MI355XB300SGLang · FP8GB200 NVL72—no exact @30 resultGB300 NVL72—no exact @30 resultQwen3.5 397B
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
B200—no exact @30 resultMI355X—no data for this scenarioB300—no exact @30 resultGB200 NVL72—no data for this scenarioGB300 NVL72Dynamo SGLang · FP4
Current cost and change versus the latest validated platform result 30–60 days earlier.
Platforms without a valid 30-day comparison show current cost only.
If a chip does not have FP4 spec decoding available, the next best available configuration is used.