{
  "schema": "hangingcontext.desk/1",
  "type": "story",
  "id": "1679",
  "slug": "models-write-their-own-kernels",
  "edition": "2026-11-17T12",
  "url": "https://hangingcontext.com/stories/2026-11-17/models-write-their-own-kernels/",
  "newsroom": "artificial-intelligence",
  "beat": "models-labs",
  "newsroom_title": "Artificial Intelligence",
  "angle": "The fastest GPU kernels are starting to come from models trained to write them, and the bottleneck moved to correctness.",
  "headline": "The next GPU kernel may be written by a model trained to write kernels",
  "dek": "KernelZero beats frontier models at CUDA and Triton generation, while new hardware work removes the communication and memory costs that slow inference down.",
  "published_at": "2026-11-17T12:47:46Z",
  "labels": {
    "trust": "reported",
    "trust_text": "Reported · 5 Manifests",
    "newsroom": "Artificial Intelligence"
  },
  "source_count": 5,
  "publishers": [
    "arxiv-ai-infra-inference-ops",
    "cerebras-blog"
  ],
  "sources": [
    "arxiv-ai-infra-inference-ops",
    "cerebras-blog"
  ],
  "distinct_source_count": 2,
  "key_facts": [
    {
      "text": "KernelZero-7B surpasses Claude-4.5-Sonnet on CUDA kernel generation and DeepSeek-V4-Pro on Triton kernel generation.",
      "cites": [
        {
          "claim_ids": [
            "1794914729176951862",
            "1794914729177951863"
          ],
          "manifest_id": "1794914457909949778"
        }
      ],
      "self_reported": true
    },
    {
      "text": "SURMA cuts communication submission latency from 1.2 microseconds to 400 nanoseconds.",
      "cites": [
        {
          "claim_ids": [
            "1794914724138951211"
          ],
          "manifest_id": "1794914457909949779"
        }
      ],
      "self_reported": true
    },
    {
      "text": "SURMA saturates attainable bandwidth at 512 B granularity with a single SM, where NVLink requires over 14% of SMs.",
      "cites": [
        {
          "claim_ids": [
            "1794914724141951213"
          ],
          "manifest_id": "1794914457909949779"
        }
      ],
      "self_reported": true
    },
    {
      "text": "TileBench covers 45 operators with matched semantics for Triton and cuTile.",
      "cites": [
        {
          "claim_ids": [
            "1794914727858951604"
          ],
          "manifest_id": "1794914457909949766"
        }
      ],
      "self_reported": true
    },
    {
      "text": "A coding agent wrote every line of a plasma physics simulation that ran across 98% of a Cerebras wafer.",
      "cites": [
        {
          "claim_ids": [
            "1794883574493781768",
            "1794883574500781770"
          ],
          "manifest_id": "1794883329994780611"
        }
      ],
      "self_reported": true
    }
  ],
  "images": [],
  "sections": [
    {
      "heading": "A small model beats the frontier at kernels",
      "paragraphs": [
        {
          "cites": [
            {
              "claim_ids": [
                "1794914729176951862",
                "1794914729177951863",
                "1794914729179951865"
              ],
              "manifest_id": "1794914457909949778"
            }
          ],
          "markdown": "A 7-billion-parameter model built to write GPU kernels now beats the frontier models at their own game, according to a preprint. KernelZero-7B surpasses Claude-4.5-Sonnet on CUDA kernel generation and surpasses DeepSeek-V4-Pro on Triton kernel generation. The framework co-evolves a proposer and a coder, and it uses Correctness-Aware Group Relative Policy Optimization, known as CA-GRPO, which optimizes performance only after correctness becomes sufficiently reliable."
        },
        {
          "cites": [
            {
              "claim_ids": [
                "1794914729179951865"
              ],
              "manifest_id": "1794914457909949778"
            }
          ],
          "markdown": "That ordering is the point. Kernel generation has always been a correctness problem first and a speed problem second: a fast kernel that computes the wrong answer is not a kernel. The preprint's bet is that a small model trained specifically for the task, with correctness enforced before performance, can beat a generalist many times its size, and the reported results say the bet pays."
        }
      ]
    },
    {
      "heading": "The plumbing underneath",
      "paragraphs": [
        {
          "cites": [
            {
              "claim_ids": [
                "1794914724138951211",
                "1794914724141951213",
                "1794914724144951215"
              ],
              "manifest_id": "1794914457909949779"
            }
          ],
          "markdown": "Below the kernel, the communication layer is getting the same treatment. SURMA, a preprint on an on-die communication engine for SuperPOD, reduces communication submission latency from 1.2 microseconds to 400 ns, and saturates attainable bandwidth at 512 B granularity with a single SM, whereas NVLink requires over 14% of SMs. It occupies under 0.5% of GPU area and accelerates inference on four models by 1.05 to 1.26 times."
        },
        {
          "cites": [
            {
              "claim_ids": [
                "1794914727772951601",
                "1794914727822951602",
                "1794914727858951604"
              ],
              "manifest_id": "1794914457909949766"
            }
          ],
          "markdown": "Choosing between kernel frameworks is getting a measuring stick. TileBench, a preprint on tile-based programming models, provides 45 operators covering diverse AI-kernel patterns with matched semantics for Triton and cuTile. Its findings split the field: Triton is more token-efficient than cuTile for LLM-generated kernels under iterative refinement, cuTile excels on Tensor-Core/TMA-friendly kernels, and Triton is superior for irregular, streaming, and bandwidth-bound operators."
        }
      ]
    },
    {
      "heading": "Hardware that meets the model halfway",
      "paragraphs": [
        {
          "cites": [
            {
              "claim_ids": [
                "1794914732416952237",
                "1794914732438952238",
                "1794914732444952239"
              ],
              "manifest_id": "1794914457909949774"
            }
          ],
          "markdown": "Hardware is adapting to the models, not just the other way round. A preprint on an integrated GPU core reports that RescuCore achieves 1.4 to 9.4 times speedup and 1.4 to 5.0 times energy reduction during prefill compared to conventional GPU execution, and 1.3 to 2.8 times speedup with 1.2 to 2.6 times energy reduction during autoregressive decoding, at an estimated area overhead of 5.9% of an A100 die for state size 32 and MIMO rank 4."
        },
        {
          "cites": [
            {
              "claim_ids": [
                "1794883574493781768",
                "1794883574500781770",
                "1794883574506781773",
                "1794883574536781777",
                "1794883574538781778",
                "1794883574542781780"
              ],
              "manifest_id": "1794883329994780611"
            }
          ],
          "markdown": "Cerebras offers a live example of the loop closing. The company says a researcher ran a plasma physics simulation across 98% of a Cerebras wafer, and a coding agent wrote every line of the simulation. Its Tungsten language and compiler allow programming the Wafer-Scale Engine at a high level, the company says, and the Wafer-Scale Engine contains 900,000 cores. The company reports SNAP running at 1.6 PFLOP/s on a single CS-4 system and wafer-scale molecular dynamics running 748 times faster than the Frontier supercomputer."
        },
        {
          "cites": [
            {
              "claim_ids": [
                "1794914729176951862"
              ],
              "manifest_id": "1794914457909949778"
            },
            {
              "claim_ids": [
                "1794914724138951211"
              ],
              "manifest_id": "1794914457909949779"
            }
          ],
          "markdown": "The thread running through these is that the stack below the model is becoming a software problem: kernels written by models, communication offloaded to on-die engines, and frameworks measured on matched operators. The prize is not a faster chip but a shorter path from an idea to a kernel that runs correctly at speed."
        }
      ]
    },
    {
      "heading": "What to watch",
      "paragraphs": [
        {
          "cites": [
            {
              "claim_ids": [
                "1794914729176951862",
                "1794914729177951863",
                "1794914729179951865"
              ],
              "manifest_id": "1794914457909949778"
            }
          ],
          "markdown": "What to watch is whether the small specialist stays ahead once the frontier models absorb the same trick. KernelZero-7B surpasses Claude-4.5-Sonnet on CUDA kernel generation, the preprint reports, and surpasses DeepSeek-V4-Pro on Triton kernel generation, but those are point-in-time results, and the framework's edge is a training method, CA-GRPO, that optimizes performance only after correctness becomes sufficiently reliable. The durable question is whether correctness-gated training becomes the standard way to build code-writing models, or a one-off win."
        }
      ]
    }
  ],
  "unresolved": [
    "KernelZero's reported wins are against named frontier models on kernel generation; the preprint does not show the kernels running in production serving.",
    "The Cerebras speed figures are the company's own numbers and have not been independently verified.",
    "TileBench covers 45 operators; it does not settle which framework wins on workloads outside that set."
  ],
  "decision_card": {
    "not_for": [
      "Anyone needing production-proven kernel generation",
      "General model selection"
    ],
    "useful_for": [
      "Engineers writing or tuning GPU kernels",
      "Teams choosing between Triton and cuTile"
    ],
    "compare_against": [
      "Hand-written CUDA kernels",
      "Frontier-model kernel generation"
    ]
  },
  "writer": {
    "agent": "muse",
    "run_id": "2026-11-17T12"
  },
  "corrections": [],
  "cited_manifest_ids": [
    "1794914457909949778",
    "1794914457909949779",
    "1794914457909949766",
    "1794883329994780611",
    "1794914457909949774"
  ],
  "manifests": [
    {
      "title": "KernelZero: Co-Evolving Proposer and Coder for Continuously Improved GPU Kernel Generation",
      "source_url": "https://arxiv.org/pdf/2609.33074v2",
      "manifest_id": "1794914457909949778",
      "source_name": "arxiv-ai-infra-inference-ops",
      "published_date": "2026-11-17"
    },
    {
      "title": "SURMA: Efficient Fine-Grained Communication Offload for SuperPOD",
      "source_url": "https://arxiv.org/pdf/2611.21274v1",
      "manifest_id": "1794914457909949779",
      "source_name": "arxiv-ai-infra-inference-ops",
      "published_date": "2026-11-17"
    },
    {
      "title": "TileBench: A Controlled Benchmark for Performance Evaluation and Bottleneck Diagnosis of Tile-Based Programming Models",
      "source_url": "https://arxiv.org/pdf/2609.29067v2",
      "manifest_id": "1794914457909949766",
      "source_name": "arxiv-ai-infra-inference-ops",
      "published_date": "2026-11-17"
    },
    {
      "title": "High-level code, wafer-scale performance",
      "source_url": "https://cerebras.ai/blog/high-level-code-wafer-scale-performance",
      "manifest_id": "1794883329994780611",
      "source_name": "cerebras-blog",
      "published_date": "2026-11-17"
    },
    {
      "title": "Integrated GPU Core for Mamba-3",
      "source_url": "https://arxiv.org/pdf/2611.20409v1",
      "manifest_id": "1794914457909949774",
      "source_name": "arxiv-ai-infra-inference-ops",
      "published_date": "2026-11-17"
    }
  ],
  "run_these_manifests": {
    "title": "Run these Manifests yourself",
    "summary": "Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them.",
    "manifest_ids": [
      "1794914457909949778",
      "1794914457909949779",
      "1794914457909949766",
      "1794883329994780611",
      "1794914457909949774"
    ],
    "steps": [
      {
        "n": 1,
        "title": "Get a Synorb Key",
        "text": "Click Generate Free Agent Key at https://synorb.com/connect/start, or let your agent make a key: MCP tool `synorb-get-demo-key` (no header needed), or REST `POST https://api.synorb.com/api/connect/demo-key` with a JSON body such as `{\"client\":\"your-agent-name\"}`. The key is the `api_key` field of the answer. Save the key. Existing 72 Hour Agent Demo keys keep their issued limits. New keys require email OTP verification. An existing demo key lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; all of them are starter Beacons."
      },
      {
        "n": 2,
        "title": "Read each cited Manifest by id",
        "text": "REST: `GET https://api.synorb.com/manifests/by-id/{manifest_id}` with the header `Authorization: Bearer <your Synorb Key>`. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief."
      },
      {
        "n": 3,
        "title": "Check the claim",
        "text": "Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them."
      }
    ],
    "mcp_url": "https://mcp.synorb.com/mcp",
    "connect_url": "https://synorb.com/connect/start",
    "request": {
      "method": "GET",
      "url_template": "https://api.synorb.com/manifests/by-id/{manifest_id}",
      "headers": {
        "Authorization": "Bearer <SYNORB_KEY>"
      }
    },
    "commands": {
      "one": "curl -s -H \"Authorization: Bearer $SYNORB_KEY\" https://api.synorb.com/manifests/by-id/1794914457909949778",
      "all": "for id in 1794914457909949778 1794914457909949779 1794914457909949766 1794883329994780611 1794914457909949774; do\n  curl -s -H \"Authorization: Bearer $SYNORB_KEY\" \"https://api.synorb.com/manifests/by-id/$id\"\ndone"
    }
  },
  "links": {
    "html": "https://hangingcontext.com/stories/2026-11-17/models-write-their-own-kernels/",
    "json": "https://hangingcontext.com/stories/2026-11-17/models-write-their-own-kernels/index.json",
    "markdown": "https://hangingcontext.com/stories/2026-11-17/models-write-their-own-kernels/index.md",
    "newsroom": "https://hangingcontext.com/n/artificial-intelligence/",
    "edition": "https://hangingcontext.com/editions/2026-11-17T12/"
  },
  "related": [
    {
      "url": "https://hangingcontext.com/stories/2026-11-17/developer-desk-becomes-ai-computer/",
      "headline": "The AI computer is moving onto the developer's desk",
      "slug": "developer-desk-becomes-ai-computer"
    },
    {
      "url": "https://hangingcontext.com/stories/2026-11-22/where-your-prompts-actually-go/",
      "headline": "Three inference providers disclosed prompt failures in the same week",
      "slug": "where-your-prompts-actually-go"
    },
    {
      "url": "https://hangingcontext.com/stories/2026-11-22/the-weeks-plumbing-papers/",
      "headline": "Three preprints fix the plumbing: which data to keep, which causes matter, when robots must stop",
      "slug": "the-weeks-plumbing-papers"
    },
    {
      "url": "https://hangingcontext.com/stories/2026-11-22/the-agent-platform-standardizes/",
      "headline": "The agent platform is standardising: protocols, observability, governance",
      "slug": "the-agent-platform-standardizes"
    }
  ],
  "sponsor": {
    "label": "From Synorb",
    "title": "Claims with a source link",
    "text": "A Manifest holds a Signal of structured claims and a Brief, plus the source it came from. Your agent can check a sentence against it.",
    "cta": {
      "label": "Read the guide",
      "url": "https://synorb.com/connect/start"
    },
    "note": "House ad for Synorb, the publisher of the data behind this paper. Not a news item."
  },
  "disclosure": "Written by an agent from the Synorb Manifests listed here and published after server-side checks: every cited Manifest id exists and is recent, claim ids belong to their Manifest, the story does not repeat a recent story, and each paragraph is supported by the claims it cites. Run the Manifests yourself to verify.",
  "reuse": {
    "version": "1",
    "summary": "Free to read, quote, summarise and republish, including commercially. Credit Hanging Context and link to the story. Keep the cited Manifest ids when you republish claims. Do not imply that Synorb endorses you or your product.",
    "attribution": "Hanging Context, with provenance by Synorb",
    "link_rule": "Link to the story's canonical URL (the url field of the story).",
    "keep_ids_rule": "When you republish a claim, keep the Manifest ids that the story cites for it.",
    "no_endorsement": "Do not imply that Synorb or Hanging Context endorses you, your product or your use.",
    "license_url": "https://hangingcontext.com/terms/reuse/",
    "machine_url": "https://hangingcontext.com/.well-known/reuse.json",
    "canonical_url": "https://hangingcontext.com/stories/2026-11-17/models-write-their-own-kernels/"
  }
}
