Artificial Analysis Publishes AA-AgentPerf Benchmarks for DeepSeek V4.1 Flash
11/07/2026 — 11/07, 06:17·1 sources (all)·3 reports (all)
Key takeaways
- Artificial Analysis has published complete AA-AgentPerf benchmark results for DeepSeek V4.1 Flash, assessing system performance when running…
Background & analysis
On 2026-11-07, AI benchmarking firm Artificial Analysis announced via its X account that it had released AA-AgentPerf results for DeepSeek V4.1 Flash. AA-AgentPerf is the firm's own benchmark for measuring how datacenter systems handle realistic agentic coding workloads. Artificial Analysis said that over the past month, a newer vLLM release together with new vLLM kernels had lifted DeepSeek V4.1 Flash's tok/s per agent on GB300 NVL72 by up to 1.5x, as measured on AA-AgentPerf.
In the same post, Artificial Analysis noted that NVIDIA had submitted updated Vera Rubin NVL72 configs for DeepSeek V4.1 Flash, and it compared these against the improved GB300 results. At 260 tok/s per agent, Vera Rubin delivered 5.0x GB300's throughput per GPU; at 80 tok/s per agent, the gap was 1.7x. In a single-agent setting, Vera Rubin reached 1,249 tok/s versus 848 tok/s on GB300, or about 1.47x.
Artificial Analysis added that the full AA-AgentPerf results for DeepSeek V4.1 Flash are now published on its platform. At this point, the speed figures, the GB300 versus Vera Rubin comparisons, and the attribution of the improvement to the newer vLLM version and kernels all come from Artificial Analysis's own statements; no separate confirmation from DeepSeek or NVIDIA has appeared in the available coverage.
AI-generated from 3 reports · updated 28 minutes ago
Latest turnArtificial Analysis has published the full AA-AgentPerf benchmark results for DeepSeek V4.1 Flash, evaluating datacenter system performance under realistic agentic coding workloads.
Related AI tools
- Heat index
- 94
- All sources
- 1
- All reports
- 3
- First seen
- 2 hours ago
Reports on this story headlines open the original
Artificial Analysis has published the full AA-AgentPerf benchmark results for DeepSeek V4.1 Flash, evaluating datacenter system performance under realistic agentic coding workloads.
X：Artificial Analysis (@artificialanlys)First-partyAI score 46According to Artificial Analysis, NVIDIA has submitted updated Vera Rubin NVL72 configurations for DeepSeek V4.1 Flash, delivering up to 5.0x the per-GPU throughput of GB300 at 260 tok/s per agent.
X：Artificial Analysis (@artificialanlys)First-partyAI score 53According to Artificial Analysis, DeepSeek V4.1 Flash speeds on GB300 NVL72 have increased by up to 1.5x per agent over the past month, driven by newer vLLM versions and optimized kernels.
X：Artificial Analysis (@artificialanlys)First-partyAI score 47
Other stories people are talking about
- Heat index 831RisingThinking Machines Lab releases Inkling 2 Small amid reported $5B funding round12 sources
- Heat index 69Creator makes a 10-minute documentary with DeepSeek in six hours1 sources
- Heat index 54DeepSeek Harness team open-sources the Super-agent architecture1 sources
- Heat index 55DeepSeek secures RMB 1.65 billion in financing from a group of Zhejiang industrialists1 sources
- Heat index 59Developer Open-Sources Xiaowei Assistant Android AI Agent1 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion