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
    $0.496About the same cost as B200
    ATOM¹ · FP4
    B300
    $0.31933% cheaper than B200
    Dynamo SGLang · FP4
    GB200 NVL72
    $2.668464% more expensive than B200
    Dynamo vLLM · FP4
    GB300 NVL72
    $0.18062% cheaper than B200
    Dynamo SGLang · FP4
    View details
  • DeepSeek V4 Pro 1.6T

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    vLLM · FP4
    MI355X
    cannot reach @100
    B300
    $0.07346% cheaper than B200
    vLLM · FP4
    GB200 NVL72
    no data for this scenario
    GB300 NVL72
    $0.03574% cheaper than B200
    Dynamo vLLM · FP4
    View details
  • Kimi K3 2.8T

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    cannot reach @100
    MI355X
    no data for this scenario
    B300
    $0.197No B200 baseline to compare against
    vLLM · FP4
    GB200 NVL72
    cannot reach @100
    GB300 NVL72
    no data for this scenario
    View details
  • MiniMax M3 428B

    8K/1K

    B200
    vLLM · FP4
    MI355X
    $0.075About the same cost as B200
    ATOM¹ · FP4
    B300
    $0.08819% more expensive than B200
    vLLM · FP4
    GB200 NVL72
    $0.295300% more expensive than B200
    Dynamo vLLM · FP8STP
    GB300 NVL72
    $0.10441% more expensive than B200
    Dynamo vLLM · FP8
    View details
  • MiniMax M3 428B

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    vLLM · FP4
    MI355X
    no data for this scenario
    B300
    no exact @100 result
    GB200 NVL72
    no data for this scenario
    GB300 NVL72
    no data for this scenario
    View details
  • GLM5.2

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    SGLang · FP4
    MI355X
    no data for this scenario
    B300
    $0.06118% cheaper than B200
    SGLang · FP4
    GB200 NVL72
    no data for this scenario
    GB300 NVL72
    no data for this scenario
    View details
  • Qwen3.5 397B

    8K/1K

    B200
    TRTLLM · FP4
    MI355X
    $0.09981% more expensive than B200
    SGLang · FP4
    B300
    $0.06111% more expensive than B200
    SGLang · FP4
    GB200 NVL72
    $0.112104% more expensive than B200
    Dynamo SGLang · FP8
    GB300 NVL72
    $0.06518% more expensive than B200
    Dynamo TRTLLM · FP4
    View details
  • Qwen3.5 397B

    Long Context Multi-Turn Realistic Agentic Scenario (AgentX)

    B200
    SGLang · FP4
    MI355X
    no data for this scenario
    B300
    $0.01145% more expensive than B200
    SGLang · FP4
    GB200 NVL72
    no data for this scenario
    GB300 NVL72
    $0.00929% more expensive than B200
    Dynamo SGLang · FP4
    View details

— = no result. ∞ = B200 baseline unavailable.

If a chip does not have FP4 spec decoding available, the next best available configuration is used.