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 SGLang · FP4 | $0.28060% more expensive than B200 SGLang · FP4 | $0.28261% more expensive than B200 SGLang · FP4 | $0.864393% more expensive than B200 Dynamo vLLM · FP4 | $0.771340% more expensive than B200 Dynamo SGLang · FP4 · STP |
DeepSeek V4 Pro 1.6TLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | vLLM · FP4 | $0.01942% more expensive than B200 MoRI SGLang · FP4 | $0.056306% more expensive than B200 SGLang · FP4 | $0.02153% more expensive than B200 Dynamo SGLang · FP4 | $0.0145% more expensive than B200 Dynamo vLLM · FP4 |
Kimi K3 2.8TLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | Dynamo vLLM · FP4 | no exact @75 result | $0.1389% more expensive than B200 vLLM · FP4 | $0.16833% more expensive than B200 Dynamo vLLM · FP4 | $0.04862% cheaper than B200 Dynamo vLLM · FP4 |
Kimi K2.5/2.6/2.7-Code 1T8K/1K | Dynamo TRTLLM · FP4 · STP | $0.1865% cheaper than B200 ATOM¹ · FP4 · STP | $0.25530% more expensive than B200 vLLM · FP4 · STP | $0.11740% cheaper than B200 Dynamo TRTLLM · FP4 · STP | $0.12338% cheaper than B200 Dynamo TRTLLM · FP4 · STP |
MiniMax M3 428B8K/1K | vLLM · FP4 | $0.06415% more expensive than B200 ATOM¹ · 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) | TRTLLM · FP4 | $0.00727% cheaper than B200 ATOM¹ · FP4 | $0.00811% cheaper than B200 TRTLLM · FP4 | $0.01233% more expensive than B200 Dynamo vLLM · FP4 | no exact @75 result |
GLM5.2Long Context Multi-Turn Realistic Agentic Scenario (AgentX) | SGLang · FP8 | $0.02361% cheaper than B200 SGLang · FP4 | no exact @75 result | no exact @75 result | no exact @75 result |
Qwen3.5 397B8K/1K | TRTLLM · FP4 | $0.06142% more expensive than B200 SGLang · FP4 | $0.045About the same cost as B200 SGLang · FP4 | $0.223415% more expensive than B200 Dynamo SGLang · FP8 · STP | $0.042About the same cost as B200 Dynamo TRTLLM · FP4 · STP |
Qwen3.5 397BLong Context Multi-Turn Realistic Agentic Scenario (AgentX) | SGLang · FP8 | $0.00652% cheaper than B200 SGLang · FP4 | $0.00840% cheaper than B200 SGLang · FP4 | $0.00652% cheaper than B200 Dynamo SGLang · FP4 | no exact @75 result |
8K/1K
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
Long Context Multi-Turn Realistic Agentic Scenario (AgentX)
8K/1K
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)
— = no result. ∞ = B200 baseline unavailable.
If a chip does not have FP4 spec decoding available, the next best available configuration is used.
