AA-AgentPerf adds DeepSeek V4.1 Flash benchmarks across NVIDIA and AMD hardware

What happened

Artificial Analysis added DeepSeek V4.1 Flash to AA-AgentPerf, its benchmark replaying real agentic coding sessions with prompt caching and tool-call delays. At 100 tokens/s per user, GB300 leads with approximately 261M tokens/s per MW, followed by B300 at approximately 220M and MI355X at approximately 128M. Compared with DeepSeek V4 Pro 0813, V4.1 Flash scores 39 versus 36 on the Artificial Analysis Intelligence Index and delivers approximately 1.3x to 6.2x the throughput per MW at that user output speed, depending on hardware.

The story so far

Artificial Analysis added DeepSeek V4.1 Flash and AMD MI355X results to its AA-AgentPerf leaderboard, reporting that NVIDIA’s GB300 led the listed V4.1 Flash systems in throughput per megawatt, while MI355X outperformed B200 under specific DeepSeek V4 Pro 0813 workloads. AA-AgentPerf replays real coding agent sessions with prompt caching and tool-call delays. For V4.1 Flash at 100 output tokens per second per user, Artificial Analysis reported approximately 261 million tokens per second per MW for GB300, 220 million for B300 and 128 million for MI355X. At that per-user speed, it reported V4.1 Flash delivered approximately 1.3 to 6.2 times V4 Pro 0813’s throughput per MW, depending on hardware. Its Intelligence Index scores were 39 for V4.1 Flash and 36 for V4 Pro 0813. In its broader AMD comparison, Artificial Analysis reported that MI355X on a single 8-GPU node delivered approximately 15% to 40% more throughput per MW than B200 on V4 Pro 0813 at 35–75 tokens per second per user. It said V4.1 Flash matched NVIDIA Blackwell performance for a single agent but fell behind as load increased; GLM-5.2/5.3 and Kimi K3 also trailed the compared NVIDIA systems. All MI355X results used ROCm nightly and development builds of vLLM and SGLang; the Kimi K3 comparisons used NVIDIA multi-node configurations.

Keep reading

Trending now

1. 1Meta Releases Muse Spark Open With Mac and Cloud Support, Topping US Open-Weight Index
2. 2Google Launches Gemini 4 Argon Globally, Expands Paid Access as Benchmark Results Vary
3. 3OpenAI Updates GPT-6 Astra Pro in ChatGPT With Parallel Agents and Faster Responses
4. 4Meta’s New Business Terms Raise Dispute Over Website Crawling and AI Training Rights