Hyperscaler cost ↓ Lower is better Source: InferenceX & SemiAnalysis Market July 2026 AI Cloud TCO Model
| Model · Scenario | B200 · Reference | MI355X | B300 | GB200 NVL72 | GB300 NVL72 |
|---|---|---|---|---|---|
DeepSeek V4 Pro 1.6T8K/1K | Dynamo vLLM · FP4 | $0.534230% more expensive than B200 SGLang · FP4 | $0.18011% more expensive than B200 Dynamo SGLang · FP4 | $0.864434% more expensive than B200 Dynamo vLLM · FP4 | $0.771376% more expensive than B200 Dynamo SGLang · FP4 · STP |
DeepSeek V4 Pro 1.6TLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | vLLM · FP4 | — cannot reach @75 | $0.04740% cheaper than B200 vLLM · FP4 | — no data for this scenario | $0.01778% cheaper than B200 Dynamo vLLM · FP4 |
Kimi K3 2.8TLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | — cannot reach @75 | — no data for this scenario | $0.177No B200 baseline to compare against vLLM · FP4 | $0.597No B200 baseline to compare against Dynamo vLLM · FP4 | — no data for this scenario |
MiniMax M3 428B8K/1K | vLLM · FP4 | $0.123122% more expensive than B200 vLLM · FP4 | $0.07434% more expensive than B200 vLLM · FP4 | $0.185234% more expensive than B200 Dynamo vLLM · FP8 · STP | $0.212281% more expensive than B200 Dynamo vLLM · FP8 · STP |
MiniMax M3 428BLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | vLLM · FP4 | — no data for this scenario | $0.012About the same cost as B200 vLLM · FP4 | — no data for this scenario | — no data for this scenario |
GLM5.2Long Context Multi-Turn Realistic Agentic Scenario (AgentX) | SGLang · FP4 | — no data for this scenario | — no exact @75 result | — no data for this scenario | — no data for this scenario |
Qwen3.5 397B8K/1K | SGLang · FP4 | $0.08254% more expensive than B200 SGLang · FP4 | $0.0507% cheaper than B200 SGLang · FP4 | $0.314488% more expensive than B200 Dynamo SGLang · FP8 · STP | $0.153187% more expensive than B200 Dynamo SGLang · FP4 · STP |
Qwen3.5 397BLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | SGLang · FP8 | — no data for this scenario | $0.00841% cheaper than B200 SGLang · FP4 | — no data for this scenario | $0.00935% cheaper than B200 Dynamo SGLang · FP4 |
DeepSeek R1 0528 671BMaintenance8K/1K | SGLang · FP4 | $0.10332% cheaper than B200 MoRI SGLang · FP4 | $0.431184% more expensive than B200 SGLang · FP8 | $0.308103% more expensive than B200 Dynamo SGLang · FP4 · STP | $1.8691,134% more expensive than B200 Dynamo SGLang · FP8 · STP |
Kimi K2.5/2.6/2.7-Code 1TDeprecated8K/1K | Dynamo vLLM · FP4 · STP | $0.207About the same cost as B200 vLLM · FP4 · STP | $0.25520% more expensive than B200 vLLM · FP4 · STP | $0.28635% more expensive than B200 Dynamo vLLM · FP4 · STP | $0.25420% more expensive than B200 Dynamo vLLM · FP4 · STP |
GLM5/5.1 744BDeprecated8K/1K | SGLang · FP4 | $0.637172% more expensive than B200 SGLang · FP8 | $0.30631% more expensive than B200 SGLang · FP4 | $0.18023% cheaper than B200 Dynamo SGLang · FP4 | $0.28924% more expensive than B200 Dynamo SGLang · FP4 |
gpt-oss 120BDeprecated8K/1K | vLLM · FP4 · STP | $0.013About the same cost as B200 vLLM · FP4 · STP | — no data for this scenario | — no data for this scenario | — no data for this scenario |
MiniMax M2.5/2.7 230BDeprecated8K/1K | Dynamo vLLM · FP4 · STP | $0.08666% more expensive than B200 vLLM · FP4 · STP | $0.06525% more expensive than B200 Dynamo vLLM · FP4 · STP | $0.0487% cheaper than B200 Dynamo vLLM · FP4 · STP | $0.06016% more expensive than B200 Dynamo vLLM · FP4 · STP |
Llama 3.3 70B InstructDeprecated8K/1K | vLLM · FP4 · STP | $0.403235% more expensive than B200 vLLM · FP4 · STP | — no data for this scenario | — no data for this scenario | — no data for this scenario |
8K/1K
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
8K/1K
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
8K/1K
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
8K/1K
8K/1K
8K/1K
8K/1K
8K/1K
8K/1K
— = no result. ∞ = B200 baseline unavailable.
If a chip does not have FP4 spec decoding available, the next best available configuration is used.
