Skip to content
← All news

News2 min read

Top 3 on every MLSys 2026 track

KDA placed #1 on MoE, #2 on DSA and #3 on GDN in the Full-Agent tracks of the MLSys 2026 FlashInfer contest. Its best kernel ran 19.1× faster than the FlashInfer baseline, and on two of the five it was still slower.

Kernels designed with KDA took a podium place on every track of the MLSys 2026 FlashInfer kernel contest, in the Full-Agent division, where no human writes kernel code: first on MoE, second on DSA, third on GDN. The team entered as HAN Lab Kernel Mafia.

Full-Agent results · mlsys26.flashinfer.ai
MoEDSAGDN#2 GEMM People#3 Insider#1 Dogacel#3 UW SyFI#1 UW SyFI#2 LLM-CUDA#1 KDA#2 KDA#3 KDA

Swipe the diagram sideways to see all of it →

1/0 Three tracks, each scored on the organisers’ machines.

A kernel contest is the closest thing to an ideal test for an agent loop. The score is a measurement, the machine belongs to the organisers, the reference implementation is theirs, and correctness is a gate rather than a matter of interpretation. A kernel that is faster because it has quietly become incorrect does not place. It fails.

The loop that wrote them ​

The agent is the Humanize coder–verifier loop: one agent plans and writes a kernel change, and an independent verifier checks it for correctness and speed before anything is kept. Two skills sit beside it. KernelWiki is a searchable store of production kernels, past submissions and documentation, and ncu-report-skill turns an Nsight Compute report into evidence an agent can act on.

The kernel agent loop
CoderVerifiera kernel changecorrect? faster? reviewKernelWikincu-report-skill

Swipe the diagram sideways to see all of it →

1/0 A coder agent plans and writes the next kernel change.

The speedups, as submitted ​

A placement ranks teams against each other; it does not say how fast the kernel is. The team's technical report gives the speed of each submitted kernel against the official FlashInfer baseline, as the arithmetic mean over the official workloads.

Submitted entries · speedup vs. the FlashInfer baseline · B200
DSA indexerDSA track
19.08×
DSA attentionDSA track
4.54×
GDN prefillGDN track
1.92×
GDN decodeGDN track
0.80×
MoE FP8MoE track
0.65×
  • KDA, as submitted
Technical report, Table 1. Below the 1.00× line, the submitted kernel was slower than the baseline.

Two of the five kernels were slower than the baseline. One of them, MoE FP8 at 0.65×, still won its track: the baseline there calls a closed-source TensorRT-LLM GEMM, and no Full-Agent entry placed ahead of it.An earlier version of this post gave 1.4× on MoE, 33.3× on DSA and 17.6× on GDN. Those figures do not match the technical report, so they are gone.

After the contest: what each piece is worth ​

After the deadline, the team reran KDA and K-Search, another agentic kernel system, on the same workloads under the same strict 48-hour budget. KDA came out ahead on all three tracks. Switch the baseline to K-Search to read the margin directly.

Post-contest rerun · 48-hour budget · speedup vs. FlashInfer
DSA
3.76×
11.90×
GDN
1.04×
1.16×
MoE FP8
0.27×
0.72×
Technical report, Table 2.

On the DSA top-k indexer, the report adds the pieces one at a time. The loop matters most, and each skill adds to it.

DSA top-k indexer · adding one piece at a time
1.37×
K-Search
3.71×
Humanize
6.14×
+ KernelWiki
8.58×
+ ncu-report-skill
  • Speedup
Technical report, Table 3. Speedup over the FlashInfer baseline.

DSA top-k indexer, full loop

8.58×

Humanize, KernelWiki and the profiler skill together, from 1.37× without them.

Best submitted kernel

19.08×

The DSA indexer as entered in the contest, against the FlashInfer baseline.

The kernels, the prompts and the report are published.

The contest kernels · prompts and report · the results · KDA

Read next

News6 KDA PRs merged into SGLangSince August, SGLang has merged six pull requests carrying kernels our agents wrote, the best adding 8.7% end-to-end throughput. The MLSys contest kernels are public with a benchmark anyone can rerun. One merge elsewhere, Flash-KMeans, was reverted the next day.Read it BlogKDA²: Kernel Design Agents (KDA) Optimize Kimi Delta Attention (KDA)Our agents wrote Kimi Delta Attention kernels that run up to 2.96× faster than FlashKDA on B300 with a tenth of its state error. Here is how, and how the agents tried to cheat along the way.Read it News6.5× faster MSA prefill indexerOn B300, KDA found the hardware the MSA prefill and decode indexers were leaving idle: a 6.5× geometric mean on prefill, up to 14× on chunk tails, up to 3.3× on decode, and bitwise-identical output throughout. Humanfia-reported.Read it