NVIDIA just posted the first agentic AI benchmark results where GB300 NVL72 runs up to 20x more coding agents per megawa…
Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-06-12
NVIDIA's GB300 NVL72 achieves up to 20x more coding agents per megawatt than the H200 in the first agentic AI benchmark results published by Artificial Analysis using their AgentPerf framework.
Appears in
Extraction
Topics: nvidia-hardwareagentic-ai-benchmarksinference-efficiencyai-hardware
Claims
- NVIDIA's GB300 NVL72 runs up to 20x more coding agents per megawatt than the H200.
- Traditional inference benchmarks measure only token generation speed after a single prompt, not agentic task performance.
- Artificial Analysis's AgentPerf benchmark evaluates a harder, multi-step agentic metric compared to prior token-throughput benchmarks.
- These are the first publicly posted agentic AI benchmark results from NVIDIA.
Key quotes
NVIDIA just posted the first agentic AI benchmark results where GB300 NVL72 runs up to 20x more coding agents per megawatt than H200.
Older inference benchmarks mostly ask how fast a system can produce tokens after one prompt. AgentPerf from Artificial Analysis, asks a harder [question].