# KernelBench Hard leaderboard

Coding benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelchoice.dev/leaderboards/benchmarks/coding-kernel-bench-hard

Every model the catalog carries a published KernelBench Hard value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Coding |
| Measure | Task success rate |
| Tasks | Hard GPU kernel coding tasks |
| Difficulty | Specialized systems programming |
| Published by | [MiniMax M3 model card](https://huggingface.co/MiniMaxAI/MiniMax-M3) |

A display-only benchmark for difficult GPU kernel implementation and optimization tasks.

## KernelBench Hard ranking

1 model with a published KernelBench Hard value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Task success rate | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [MiniMax M3](https://modelchoice.dev/models/minimax-m3) | MiniMax | 28.8 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
