Inference Cost per Million Tokens

Hyperscaler cost ↓ Lower is better Source: InferenceX & SemiAnalysis Market July 2026 AI Cloud TCO Model

  • DeepSeek V4 Pro 1.6T

    8K/1K

    B200
    TRTLLM · FP4
    MI355X
    cannot reach @200
    B300
    TRTLLM · FP4
    GB200 NVL72
    Dynamo SGLang · FP4
    GB300 NVL72
    Dynamo TRTLLM · FP4
  • DeepSeek V4 Pro 1.6T

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    cannot reach @200
    MI355X
    cannot reach @200
    B300
    cannot reach @200
    GB200 NVL72
    no data for this scenario
    GB300 NVL72
    cannot reach @200
  • Kimi K3 2.8T

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    cannot reach @200
    MI355X
    no data for this scenario
    B300
    cannot reach @200
    GB200 NVL72
    cannot reach @200
    GB300 NVL72
    no data for this scenario
  • MiniMax M3 428B

    8K/1K

    B200
    $0.13028% cheaper than this platform’s Jul 6 result
    vLLM · FP4
    Compare curves
    MI355X
    ATOM¹ · FP4
    B300
    $0.15617% cheaper than this platform’s Jul 6 result
    Dynamo vLLM · FP4
    Compare curves
    GB200 NVL72
    cannot reach @200
    GB300 NVL72
    Dynamo vLLM · FP8
  • MiniMax M3 428B

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    vLLM · FP4
    MI355X
    no data for this scenario
    B300
    cannot reach @200
    GB200 NVL72
    no data for this scenario
    GB300 NVL72
    no data for this scenario
  • GLM5.2

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    SGLang · FP4
    MI355X
    no data for this scenario
    B300
    cannot reach @200
    GB200 NVL72
    no data for this scenario
    GB300 NVL72
    no data for this scenario
  • Qwen3.5 397B

    8K/1K

    B200
    $0.106About the same cost as this platform’s Jul 5 result
    SGLang · FP4
    Compare curves
    MI355X
    ATOM¹ · FP8
    B300
    $0.1198% cheaper than this platform’s Jul 4 result
    SGLang · FP4
    Compare curves
    GB200 NVL72
    Dynamo SGLang · FP8
    GB300 NVL72
    $0.09894% cheaper than this platform’s Jun 25 result
    Dynamo TRTLLM · FP4
    Compare curves
  • Qwen3.5 397B

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    SGLang · FP8
    MI355X
    no data for this scenario
    B300
    SGLang · FP4
    GB200 NVL72
    no data for this scenario
    GB300 NVL72
    Dynamo 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.