{
  "api_version": "1",
  "updated": "2026-11-05",
  "latest_event": "2026-11-04",
  "rights": "© Frontier Log, all rights reserved.",
  "model": {
    "slug": "hy3",
    "name": "Hy3",
    "lab": "tencent",
    "date": "2026-07-06",
    "source": "https://hy.tencent.ai/research/hy3",
    "source_title": "Introducing Hy3",
    "flagship": false,
    "event": "https://frontierlog.ai/event/2026-07-06-hy3",
    "releases": [
      "https://frontierlog.ai/event/2026-04-23-hy3-preview"
    ],
    "variant_of": null,
    "made_by": "Tencent",
    "url": "https://frontierlog.ai/model/hy3",
    "results": [
      {
        "id": "hy3:swe-bench-verified:default:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "swe-bench-verified",
        "benchmark_name": "SWE-bench Verified",
        "setting": "default",
        "setting_name": "",
        "score": 78.0,
        "unit": "%",
        "printed": "78.0",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'SWE-bench Verified', column 'Hy3'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "SWE-agent scaffold; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "78.0",
            "where": "Model card, Appendix image, row SWE-bench Verified, column Hy3. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:swe-bench-verified:default:2",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "swe-bench-verified",
        "benchmark_name": "SWE-bench Verified",
        "setting": "default",
        "setting_name": "",
        "score": 74.4,
        "unit": "%",
        "printed": "74.4",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3 preview",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'SWE-bench Verified', column 'Hy3 preview'",
        "own": true,
        "primary": false,
        "parallel": false,
        "notes": "Hy3 preview, re-evaluated at Hy3 launch; SWE-agent scaffold; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "74.4",
            "where": "Model card, Appendix image, row SWE-bench Verified, column Hy3 preview. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          },
          {
            "url": "https://hy.tencent.ai/research/hy3-preview",
            "name": "Tencent releases Hy3 preview",
            "title": "Hy3 preview: The First Step in Rebuilding the Hy model",
            "lab": "tencent",
            "date": "2026-04-23",
            "printed": "74.4",
            "where": "Appendix table image, row SWE-bench Verified (setting -), column Hy3 preview. Hy3 preview launch page (Apr 23, 2026), appendix table, 'Hy3 preview' column. Notes: competitors at high-think; 'For the SWE-Bench series ... SWE-agent scaffold, with the best results obtained in non-thinking mode.'"
          },
          {
            "url": "https://huggingface.co/tencent/Hy3-preview",
            "name": "Tencent releases Hy3 preview card",
            "title": "tencent/Hy3-preview model card",
            "lab": "tencent",
            "date": "2026-04-23",
            "printed": "74.4%",
            "where": "Hy3-preview model card, image bench_agent_overview_v3.jpg, SWE-bench Verified panel, point Hy3 preview 74.4% (2026-04)."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:swe-bench-pro:default:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "swe-bench-pro",
        "benchmark_name": "SWE-bench Pro",
        "setting": "default",
        "setting_name": "",
        "score": 57.9,
        "unit": "%",
        "printed": "57.9",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'SWE-bench Pro', column 'Hy3'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "SWE-agent scaffold; highest reasoning setting",
        "caveat": "The source doesn't say whether this is the public or the commercial set",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "57.9",
            "where": "Model card, Appendix image, row SWE-bench Pro, column Hy3. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          },
          {
            "url": "https://hy.tencent.ai/research/hy4-preview",
            "name": "Tencent releases Hy4 preview",
            "title": "Introducing Hy4 preview",
            "lab": "tencent",
            "date": "2026-08-28",
            "printed": "57.9",
            "where": "Appendix: Benchmark Results image, row SWE-bench Pro, column Hy3 (swe-agent scaffold, highest reasoning setting). Hy4 preview appendix (Aug 28, 2026), Hy3 column. Notes say 'Certain Hy3 benchmark scores may differ from previously reported results following updates to the evaluation harness, judge model, and anti-hacking mechanisms.' Same image bytes as huggingface.co/tencent/Hy4-preview assets/benchmark-appendix.jpg and github.com/Tencent-Hunyuan/Hy4-preview."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:swe-bench-pro:default:2",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "swe-bench-pro",
        "benchmark_name": "SWE-bench Pro",
        "setting": "default",
        "setting_name": "",
        "score": 46.0,
        "unit": "%",
        "printed": "46.0",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3 preview",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'SWE-bench Pro', column 'Hy3 preview'",
        "own": true,
        "primary": false,
        "parallel": false,
        "notes": "Hy3 preview, re-evaluated at Hy3 launch; SWE-agent scaffold; highest reasoning setting",
        "caveat": "The source doesn't say whether this is the public or the commercial set",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "46.0",
            "where": "Model card, Appendix image, row SWE-bench Pro, column Hy3 preview. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          },
          {
            "url": "https://hy.tencent.ai/research/hy3-preview",
            "name": "Tencent releases Hy3 preview",
            "title": "Hy3 preview: The First Step in Rebuilding the Hy model",
            "lab": "tencent",
            "date": "2026-04-23",
            "printed": "46.0",
            "where": "Appendix table image, row SWE-bench Pro, column Hy3 preview. Hy3 preview launch page (Apr 23, 2026), appendix table, 'Hy3 preview' column. Notes: competitors at high-think; 'For the SWE-Bench series ... SWE-agent scaffold, with the best results obtained in non-thinking mode.'"
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:terminal-bench-2-1:default:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "terminal-bench-2-1",
        "benchmark_name": "Terminal-Bench 2.1",
        "setting": "default",
        "setting_name": "",
        "score": 71.7,
        "unit": "%",
        "printed": "71.7",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'Terminal-Bench 2.1', column 'Hy3'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "Terminus-2 harness; 4h agent timeout; up to 500 episodes; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "71.7",
            "where": "Model card, Appendix image, row Terminal-Bench 2.1, column Hy3. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:terminal-bench-2-1:default:2",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "terminal-bench-2-1",
        "benchmark_name": "Terminal-Bench 2.1",
        "setting": "default",
        "setting_name": "",
        "score": 58.0,
        "unit": "%",
        "printed": "58.0",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3 preview",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'Terminal-Bench 2.1', column 'Hy3 preview'",
        "own": true,
        "primary": false,
        "parallel": false,
        "notes": "Hy3 preview, re-evaluated at Hy3 launch; Terminus-2 harness; 4h agent timeout; up to 500 episodes; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "58.0",
            "where": "Model card, Appendix image, row Terminal-Bench 2.1, column Hy3 preview. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management. (The Hy3 preview launch page prints only Terminal-Bench 2.0, 54.4, a different benchmark, so it is not listed.)"
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:browsecomp:default:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "browsecomp",
        "benchmark_name": "BrowseComp",
        "setting": "default",
        "setting_name": "",
        "score": 84.2,
        "unit": "%",
        "printed": "84.2",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'BrowseComp', column 'Hy3'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "internal harness; self-summary context management; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "84.2",
            "where": "Model card, Appendix image, row BrowseComp, column Hy3. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:browsecomp:default:2",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "browsecomp",
        "benchmark_name": "BrowseComp",
        "setting": "default",
        "setting_name": "",
        "score": 67.1,
        "unit": "%",
        "printed": "67.1",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3 preview",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'BrowseComp', column 'Hy3 preview'",
        "own": true,
        "primary": false,
        "parallel": false,
        "notes": "Hy3 preview, re-evaluated at Hy3 launch; internal harness; self-summary context management; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "67.1",
            "where": "Model card, Appendix image, row BrowseComp, column Hy3 preview. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          },
          {
            "url": "https://hy.tencent.ai/research/hy3-preview",
            "name": "Tencent releases Hy3 preview",
            "title": "Hy3 preview: The First Step in Rebuilding the Hy model",
            "lab": "tencent",
            "date": "2026-04-23",
            "printed": "67.1",
            "where": "Appendix table image, row BrowseComp (setting: ctx mgmt), column Hy3 preview. Hy3 preview launch page (Apr 23, 2026), appendix table, 'Hy3 preview' column. Notes: competitors at high-think; 'For the SWE-Bench series ... SWE-agent scaffold, with the best results obtained in non-thinking mode.'"
          },
          {
            "url": "https://huggingface.co/tencent/Hy3-preview",
            "name": "Tencent releases Hy3 preview card",
            "title": "tencent/Hy3-preview model card",
            "lab": "tencent",
            "date": "2026-04-23",
            "printed": "67.1%",
            "where": "Hy3-preview model card, image bench_agent_overview_v3.jpg, BrowseComp panel, point Hy3 preview 67.1%."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:mcp-atlas:default:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "mcp-atlas",
        "benchmark_name": "MCP Atlas",
        "setting": "default",
        "setting_name": "",
        "score": 79.1,
        "unit": "%",
        "printed": "79.1",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'MCP atlas (public)', column 'Hy3'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "500-task public set; 100-tool-call budget; Gemini 2.5 Pro judge; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "79.1",
            "where": "Model card, Appendix image, row MCP atlas (public), column Hy3. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:mcp-atlas:default:2",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "mcp-atlas",
        "benchmark_name": "MCP Atlas",
        "setting": "default",
        "setting_name": "",
        "score": 66.1,
        "unit": "%",
        "printed": "66.1",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3 preview",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'MCP atlas (public)', column 'Hy3 preview'",
        "own": true,
        "primary": false,
        "parallel": false,
        "notes": "Hy3 preview, re-evaluated at Hy3 launch; 500-task public set; 100-tool-call budget; Gemini 2.5 Pro judge; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "66.1",
            "where": "Model card, Appendix image, row MCP atlas (public), column Hy3 preview. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:hle:with-tools:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "hle",
        "benchmark_name": "Humanity's Last Exam",
        "setting": "with-tools",
        "setting_name": "With tools",
        "score": 53.2,
        "unit": "%",
        "printed": "53.2",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'HLE (with tools, text-only)', column 'Hy3'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "text-only subset; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "53.2",
            "where": "Model card, Appendix image, row HLE (with tools, text-only), column Hy3. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:hle:with-tools:2",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "hle",
        "benchmark_name": "Humanity's Last Exam",
        "setting": "with-tools",
        "setting_name": "With tools",
        "score": 35.4,
        "unit": "%",
        "printed": "35.4",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3 preview",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'HLE (with tools, text-only)', column 'Hy3 preview'",
        "own": true,
        "primary": false,
        "parallel": false,
        "notes": "Hy3 preview, re-evaluated at Hy3 launch; text-only subset; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "35.4",
            "where": "Model card, Appendix image, row HLE (with tools, text-only), column Hy3 preview. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:gpqa-diamond:default:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "gpqa-diamond",
        "benchmark_name": "GPQA Diamond",
        "setting": "default",
        "setting_name": "",
        "score": 90.4,
        "unit": "%",
        "printed": "90.4",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'GPQA Diamond', column 'Hy3'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "90.4",
            "where": "Model card, Appendix image, row GPQA Diamond, column Hy3. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:gpqa-diamond:default:2",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "gpqa-diamond",
        "benchmark_name": "GPQA Diamond",
        "setting": "default",
        "setting_name": "",
        "score": 87.2,
        "unit": "%",
        "printed": "87.2",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3 preview",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'GPQA Diamond', column 'Hy3 preview'",
        "own": true,
        "primary": false,
        "parallel": false,
        "notes": "Hy3 preview, re-evaluated at Hy3 launch; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "87.2",
            "where": "Model card, Appendix image, row GPQA Diamond, column Hy3 preview. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          },
          {
            "url": "https://hy.tencent.ai/research/hy3-preview",
            "name": "Tencent releases Hy3 preview",
            "title": "Hy3 preview: The First Step in Rebuilding the Hy model",
            "lab": "tencent",
            "date": "2026-04-23",
            "printed": "87.2",
            "where": "Appendix table image, row GPQA-Diamond, column Hy3 preview. Also the page's STEM bar chart, GPQA-Diamond panel, Hy3 preview 87.2 (sha256 1f70fbce5d9945de7b9eb54d0560b1f1bb66f3a067493238b680506c53b8bcb3). Hy3 preview launch page (Apr 23, 2026), appendix table, 'Hy3 preview' column. Notes: competitors at high-think; 'For the SWE-Bench series ... SWE-agent scaffold, with the best results obtained in non-thinking mode.'"
          },
          {
            "url": "https://huggingface.co/tencent/Hy3-preview",
            "name": "Tencent releases Hy3 preview card",
            "title": "tencent/Hy3-preview model card",
            "lab": "tencent",
            "date": "2026-04-23",
            "printed": "87.2",
            "where": "Hy3-preview model card, image bench_stem.jpg, GPQA-Diamond panel, Hy3 preview bar 87.2."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:hle:no-tools:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "hle",
        "benchmark_name": "Humanity's Last Exam",
        "setting": "no-tools",
        "setting_name": "No tools",
        "score": 37.0,
        "unit": "%",
        "printed": "37.0",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'HLE (no tools, text-only)', column 'Hy3'",
        "own": true,
        "primary": true,
        "parallel": false,
        "notes": "text-only subset; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "37.0",
            "where": "Model card, Appendix image, row HLE (no tools, text-only), column Hy3. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:hle:no-tools:2",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "hle",
        "benchmark_name": "Humanity's Last Exam",
        "setting": "no-tools",
        "setting_name": "No tools",
        "score": 30.0,
        "unit": "%",
        "printed": "30.0",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3 preview",
        "source": "https://hy.tencent.ai/research/hy3",
        "source_title": "Introducing Hy3",
        "source_date": "2026-07-06",
        "where": "'Appendix: Benchmark Results' table image, row 'HLE (no tools, text-only)', column 'Hy3 preview'",
        "own": true,
        "primary": false,
        "parallel": false,
        "notes": "Hy3 preview, re-evaluated at Hy3 launch; text-only subset; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "corroborated",
        "corroborated": true,
        "corroborated_on": "2026-11-05",
        "corroborated_by": [
          {
            "url": "https://huggingface.co/tencent/Hy3",
            "name": "Hy3 card",
            "title": "tencent/Hy3 model card",
            "lab": "tencent",
            "date": "2026-07-06",
            "printed": "30.0",
            "where": "Model card, Appendix image, row HLE (no tools, text-only), column Hy3 preview. Same table as the Hy3 page's appendix, but a different image file (the page's is d1af768b…). The card's top bar chart has the same bytes as the page's chart, so it is not cited. Mirrored with identical bytes on huggingface.co/tencent/Hy3-FP8 and github.com/Tencent-Hunyuan/Hy3 (assets/benchmark-appendix.png). Notes: highest reasoning tier; SWE-agent scaffold; Terminus-2, 4h timeout, 500 episodes; MCP-Atlas 500-task public set, 100-call budget, Gemini 2.5 Pro judge; BrowseComp self-summary context management."
          },
          {
            "url": "https://hy.tencent.ai/research/hy3-preview",
            "name": "Tencent releases Hy3 preview",
            "title": "Hy3 preview: The First Step in Rebuilding the Hy model",
            "lab": "tencent",
            "date": "2026-04-23",
            "printed": "30.0",
            "where": "Appendix table image, row HLE (setting: no tools, text-only), column Hy3 preview. Also the page's STEM bar chart, HLE panel, Hy3 preview 30.0 (sha256 1f70fbce5d9945de7b9eb54d0560b1f1bb66f3a067493238b680506c53b8bcb3). Hy3 preview launch page (Apr 23, 2026), appendix table, 'Hy3 preview' column. Notes: competitors at high-think; 'For the SWE-Bench series ... SWE-agent scaffold, with the best results obtained in non-thinking mode.'"
          },
          {
            "url": "https://huggingface.co/tencent/Hy3-preview",
            "name": "Tencent releases Hy3 preview card",
            "title": "tencent/Hy3-preview model card",
            "lab": "tencent",
            "date": "2026-04-23",
            "printed": "30.0",
            "where": "Hy3-preview model card, image bench_stem.jpg, HLE panel, Hy3 preview bar 30.0. The caption says domestic models were run on the text-only subset."
          }
        ],
        "source_image": "https://frontierlog.ai/sources/d1af768b6734a711.png",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026070622113393_004bbe5190853d7c35825b9257672dde.png?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1783347098;2098707098&q-key-time=1783347098;2098707098&q-header-list=&q-url-param-list=&q-signature=507857321cef2b9fef5e4ef8a8f9ab8937d2480b"
      },
      {
        "id": "hy3:terminal-bench-2-1:default:3",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "terminal-bench-2-1",
        "benchmark_name": "Terminal-Bench 2.1",
        "setting": "default",
        "setting_name": "",
        "score": 70.8,
        "unit": "%",
        "printed": "70.8",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy4-preview",
        "source_title": "Introducing Hy4 preview",
        "source_date": "2026-08-28",
        "where": "later report: Hy4 preview post, 'Appendix: Benchmark Results' table image, row 'Terminal-Bench 2.1', column 'Hy3'",
        "own": false,
        "primary": false,
        "parallel": false,
        "notes": "re-evaluated at Hy4 preview launch; Claude Code harness; up to 500 turns and a 12h timeout per trial; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "double_entry",
        "corroborated": false,
        "corroborated_on": null,
        "corroborated_by": [],
        "source_image": "https://frontierlog.ai/sources/02c751bf51c52c52.jpg",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026082812342438_ecb98509e92e4be02a40ca358414133a.jpg?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1787891670;2103251670&q-key-time=1787891670;2103251670&q-header-list=&q-url-param-list=&q-signature=b6d0c3c4d6d7e0e5219d177ec730735185ccdd1f"
      },
      {
        "id": "hy3:cybergym:default:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "cybergym",
        "benchmark_name": "CyberGym",
        "setting": "default",
        "setting_name": "",
        "score": 51.8,
        "unit": "%",
        "printed": "51.8",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy4-preview",
        "source_title": "Introducing Hy4 preview",
        "source_date": "2026-08-28",
        "where": "later report: Hy4 preview post, 'Appendix: Benchmark Results' table image, row 'CyberGym', column 'Hy3'",
        "own": false,
        "primary": true,
        "parallel": false,
        "notes": "harness not stated; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "double_entry",
        "corroborated": false,
        "corroborated_on": null,
        "corroborated_by": [],
        "source_image": "https://frontierlog.ai/sources/02c751bf51c52c52.jpg",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026082812342438_ecb98509e92e4be02a40ca358414133a.jpg?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1787891670;2103251670&q-key-time=1787891670;2103251670&q-header-list=&q-url-param-list=&q-signature=b6d0c3c4d6d7e0e5219d177ec730735185ccdd1f"
      },
      {
        "id": "hy3:mcp-atlas:default:3",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "mcp-atlas",
        "benchmark_name": "MCP Atlas",
        "setting": "default",
        "setting_name": "",
        "score": 75.0,
        "unit": "%",
        "printed": "75.0",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy4-preview",
        "source_title": "Introducing Hy4 preview",
        "source_date": "2026-08-28",
        "where": "later report: Hy4 preview post, 'Appendix: Benchmark Results' table image, row 'MCP-Atlas (public)', column 'Hy3'",
        "own": false,
        "primary": false,
        "parallel": false,
        "notes": "re-evaluated at Hy4 preview launch; 500-task public set; 100-tool-call budget; Gemini 3.1 Pro Preview judge; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "double_entry",
        "corroborated": false,
        "corroborated_on": null,
        "corroborated_by": [],
        "source_image": "https://frontierlog.ai/sources/02c751bf51c52c52.jpg",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026082812342438_ecb98509e92e4be02a40ca358414133a.jpg?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1787891670;2103251670&q-key-time=1787891670;2103251670&q-header-list=&q-url-param-list=&q-signature=b6d0c3c4d6d7e0e5219d177ec730735185ccdd1f"
      },
      {
        "id": "hy3:automationbench:default:1",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "automationbench",
        "benchmark_name": "AutomationBench",
        "setting": "default",
        "setting_name": "",
        "score": 16.1,
        "unit": "%",
        "printed": "16.1",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy4-preview",
        "source_title": "Introducing Hy4 preview",
        "source_date": "2026-08-28",
        "where": "later report: Hy4 preview post, 'Appendix: Benchmark Results' table image, row 'AutomationBench (v1.0.6)', column 'Hy3'",
        "own": false,
        "primary": true,
        "parallel": false,
        "notes": "AutomationBench v1.0.6; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "double_entry",
        "corroborated": false,
        "corroborated_on": null,
        "corroborated_by": [],
        "source_image": "https://frontierlog.ai/sources/02c751bf51c52c52.jpg",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026082812342438_ecb98509e92e4be02a40ca358414133a.jpg?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1787891670;2103251670&q-key-time=1787891670;2103251670&q-header-list=&q-url-param-list=&q-signature=b6d0c3c4d6d7e0e5219d177ec730735185ccdd1f"
      },
      {
        "id": "hy3:hle:with-tools:3",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "hle",
        "benchmark_name": "Humanity's Last Exam",
        "setting": "with-tools",
        "setting_name": "With tools",
        "score": 51.9,
        "unit": "%",
        "printed": "51.9",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy4-preview",
        "source_title": "Introducing Hy4 preview",
        "source_date": "2026-08-28",
        "where": "later report: Hy4 preview post, 'Appendix: Benchmark Results' table image, row 'HLE (with tools, text-only)', column 'Hy3'",
        "own": false,
        "primary": false,
        "parallel": false,
        "notes": "re-evaluated at Hy4 preview launch; text-only subset; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "double_entry",
        "corroborated": false,
        "corroborated_on": null,
        "corroborated_by": [],
        "source_image": "https://frontierlog.ai/sources/02c751bf51c52c52.jpg",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026082812342438_ecb98509e92e4be02a40ca358414133a.jpg?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1787891670;2103251670&q-key-time=1787891670;2103251670&q-header-list=&q-url-param-list=&q-signature=b6d0c3c4d6d7e0e5219d177ec730735185ccdd1f"
      },
      {
        "id": "hy3:gpqa-diamond:default:3",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "gpqa-diamond",
        "benchmark_name": "GPQA Diamond",
        "setting": "default",
        "setting_name": "",
        "score": 90.9,
        "unit": "%",
        "printed": "90.9",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy4-preview",
        "source_title": "Introducing Hy4 preview",
        "source_date": "2026-08-28",
        "where": "later report: Hy4 preview post, 'Appendix: Benchmark Results' table image, row 'GPQA Diamond', column 'Hy3'",
        "own": false,
        "primary": false,
        "parallel": false,
        "notes": "re-evaluated at Hy4 preview launch; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "double_entry",
        "corroborated": false,
        "corroborated_on": null,
        "corroborated_by": [],
        "source_image": "https://frontierlog.ai/sources/02c751bf51c52c52.jpg",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026082812342438_ecb98509e92e4be02a40ca358414133a.jpg?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1787891670;2103251670&q-key-time=1787891670;2103251670&q-header-list=&q-url-param-list=&q-signature=b6d0c3c4d6d7e0e5219d177ec730735185ccdd1f"
      },
      {
        "id": "hy3:hle:no-tools:3",
        "model": "hy3",
        "model_name": "Hy3",
        "model_date": "2026-07-06",
        "event": "https://frontierlog.ai/event/2026-07-06-hy3",
        "benchmark": "hle",
        "benchmark_name": "Humanity's Last Exam",
        "setting": "no-tools",
        "setting_name": "No tools",
        "score": 34.4,
        "unit": "%",
        "printed": "34.4",
        "reported_by": "tencent",
        "run_by": null,
        "label": "Hy3",
        "source": "https://hy.tencent.ai/research/hy4-preview",
        "source_title": "Introducing Hy4 preview",
        "source_date": "2026-08-28",
        "where": "later report: Hy4 preview post, 'Appendix: Benchmark Results' table image, row 'HLE (no tools, text-only)', column 'Hy3'",
        "own": false,
        "primary": false,
        "parallel": false,
        "notes": "re-evaluated at Hy4 preview launch; text-only subset; highest reasoning setting",
        "caveat": "",
        "human_verified": false,
        "human_verified_on": null,
        "check": "double_entry",
        "corroborated": false,
        "corroborated_on": null,
        "corroborated_by": [],
        "source_image": "https://frontierlog.ai/sources/02c751bf51c52c52.jpg",
        "source_image_original": "https://hy-model-ap-prod-1258344703.cos.accelerate.myqcloud.com/llm-blog/default/9a5ea02f/2026082812342438_ecb98509e92e4be02a40ca358414133a.jpg?q-sign-algorithm=sha1&q-ak=AKIDRl074nOsGdJ9zjMsCRWP3ShmgS3VtX4S&q-sign-time=1787891670;2103251670&q-key-time=1787891670;2103251670&q-header-list=&q-url-param-list=&q-signature=b6d0c3c4d6d7e0e5219d177ec730735185ccdd1f"
      }
    ],
    "price": [
      {
        "from": "2026-07-06",
        "listed": false,
        "currency": "CNY",
        "input_per_1m": 1.0,
        "cached_input_per_1m": 0.25,
        "cached_input_derived": false,
        "output_per_1m": 4.0,
        "cache": {
          "kind": "price"
        },
        "tiers": [],
        "printed": {
          "input": "¥1",
          "cached_input": "¥0.25",
          "output": "¥4"
        },
        "printed_unit": "1M tokens",
        "notes": "",
        "sources": [
          "https://hy.tencent.ai/research/hy3",
          "https://cloud.tencent.com/document/product/1823/130055",
          "https://cloud.tencent.com/product/tokenhub"
        ],
        "check": "corroborated"
      }
    ],
    "context": null,
    "prices": [
      {
        "from": "2026-07-06",
        "listed": false,
        "currency": "CNY",
        "input_per_1m": 1.0,
        "cached_input_per_1m": 0.25,
        "cached_input_derived": false,
        "output_per_1m": 4.0,
        "cache": {
          "kind": "price"
        },
        "tiers": [],
        "printed": {
          "input": "¥1",
          "cached_input": "¥0.25",
          "output": "¥4"
        },
        "printed_unit": "1M tokens",
        "notes": "",
        "sources": [
          "https://hy.tencent.ai/research/hy3",
          "https://cloud.tencent.com/document/product/1823/130055",
          "https://cloud.tencent.com/product/tokenhub"
        ],
        "check": "corroborated"
      }
    ],
    "context_history": [],
    "stories": []
  }
}