@daniel_koss · Long NBIS

Artificial Analysis just measured AI agent workloads. Tokens per second, per megawatt: NVIDIA GB300: ~261M B300: ~220M AMD MI355X: ~128M 2x the output on the best stack. Microsoft sized its new Maia 300 chip the same way, in tokens per megawatt. Every chip generation gets

Artificial Analysis@ArtificialAnlys

We’ve added DeepSeek V4.1 Flash to AA-AgentPerf, with results for NVIDIA and AMD hardware. At 100 tokens/s per user, NVIDIA GB300 is in first place with throughput up to 261M tokens/s per MW. AA-AgentPerf is our agentic inference performance benchmark. It replays real agentic

Read asLong NBIS

Since the call—

unrankedThe analyst's rank today

Medium · 0.86Confidence

Wed, Oct 14Tweeted

Thu, Oct 15 closeCounted from

2% of the basket · day 1fades out by Nov 25

Rank per close

—

Oct 15