{
  "data": {
    "similar": [
      {
        "grade": "BB",
        "json": "https://www.anchorterminal.com/tools/qwen3-embedding-8b.json",
        "name": "Qwen3-Embedding-8B",
        "score": 74.7,
        "shared": [
          "model.open-weights"
        ],
        "slug": "qwen3-embedding-8b"
      },
      {
        "grade": "BB",
        "json": "https://www.anchorterminal.com/tools/gemma-4-31b-it.json",
        "name": "Gemma 4 31B IT",
        "score": 74.1,
        "shared": [
          "model.open-weights"
        ],
        "slug": "gemma-4-31b-it"
      },
      {
        "grade": "BB",
        "json": "https://www.anchorterminal.com/tools/qwen3-embedding-4b.json",
        "name": "Qwen3-Embedding-4B",
        "score": 72.9,
        "shared": [
          "model.open-weights"
        ],
        "slug": "qwen3-embedding-4b"
      },
      {
        "grade": "BB",
        "json": "https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.json",
        "name": "Gemma 4 26B A4B IT",
        "score": 72.2,
        "shared": [
          "model.open-weights"
        ],
        "slug": "gemma-4-26b-a4b-it"
      },
      {
        "grade": "BB",
        "json": "https://www.anchorterminal.com/tools/gemma-4-12b-it.json",
        "name": "Gemma 4 12B IT",
        "score": 71.3,
        "shared": [
          "model.open-weights"
        ],
        "slug": "gemma-4-12b-it"
      },
      {
        "grade": "BB",
        "json": "https://www.anchorterminal.com/tools/gpt-oss-120b.json",
        "name": "gpt-oss-120b",
        "score": 70.2,
        "shared": [
          "model.open-weights"
        ],
        "slug": "gpt-oss-120b"
      }
    ],
    "tool": {
      "slug": "kronos",
      "name": "Kronos",
      "vendor": "Yu Shi (shiyu-coder, NeoQuasar), Tsinghua University",
      "vendorUrl": "https://github.com/shiyu-coder/Kronos",
      "kind": "weights",
      "category": "open-weights",
      "summary": "Kronos is an open-weight family of small forecasting models for financial candlestick (K-line) series, from Tsinghua University researchers. It reads open, high, low, close, volume and amount bars with timestamps and generates forecast bars, under MIT.",
      "url": "https://www.anchorterminal.com/tools/kronos",
      "markdownUrl": "https://www.anchorterminal.com/tools/kronos.md",
      "slimMarkdownUrl": "https://www.anchorterminal.com/tools/kronos.min.md",
      "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/kronos.json",
      "repo": "https://github.com/shiyu-coder/Kronos",
      "license": "MIT",
      "transports": [],
      "packages": [
        {
          "registry": "huggingface",
          "name": "NeoQuasar/Kronos-mini"
        },
        {
          "registry": "huggingface",
          "name": "NeoQuasar/Kronos-small"
        },
        {
          "registry": "huggingface",
          "name": "NeoQuasar/Kronos-base"
        },
        {
          "registry": "huggingface",
          "name": "NeoQuasar/Kronos-Tokenizer-base"
        },
        {
          "registry": "huggingface",
          "name": "NeoQuasar/Kronos-Tokenizer-2k"
        }
      ],
      "auth": "none",
      "authNotes": "The weights download from Hugging Face with no account, token or terms to accept. The Hub reports gated false for all five NeoQuasar repositories, and the code is in a public GitHub repository (checked 2026-10-25).",
      "pricing": "free",
      "pricingNotes": "Free to download and use, commercial use included, under the MIT licence named in each card's metadata and in the GitHub repository's LICENSE. Nobody sells per-call access and no inference provider serves it. Running it costs the hardware, 0.03 GB for Kronos-mini with its tokeniser and 0.43 GB for Kronos-base with its tokeniser, in FP32 (checked 2026-10-25).",
      "priceSummary": "Free",
      "where": "local",
      "x402": {
        "level": "no",
        "evidence": "Weights you download and run yourself, so there is nothing to pay per call and no payment protocol (checked 2026-10-25).",
        "endpoints": []
      },
      "toolCount": null,
      "popularity": {
        "githubStars": 43000,
        "npmWeekly": null,
        "pypiWeekly": null,
        "asOf": "2026-10-25"
      },
      "docsUrl": "https://github.com/shiyu-coder/Kronos",
      "capabilities": [
        "model.open-weights"
      ],
      "tags": [
        "open-weights",
        "model",
        "time-series",
        "forecasting",
        "finance",
        "candlestick",
        "k-line",
        "mit",
        "safetensors",
        "self-hosted"
      ],
      "lastRelease": "2026-04-13",
      "graded": true,
      "anchor": {
        "graded": true,
        "score": 55.2,
        "grade": "C",
        "agentReady": false,
        "rank": 0,
        "ranked": false,
        "notRankedWhy": "An open-weight model, graded by the open-weight checklist and ranked among open-weight models, not against tools",
        "rankOf": 1210,
        "categoryRank": 68,
        "methodology": "0.6",
        "run": "2026-10-01",
        "scores": {
          "interface": 60,
          "reach": 55,
          "risk": 50,
          "upkeep": 61
        },
        "pending": [
          "performance",
          "tasks"
        ],
        "breakdown": [
          {
            "key": "capability",
            "name": "Capability",
            "weight": 35,
            "effectiveWeight": 0,
            "pending": true,
            "points": 0,
            "reason": "Pending for this listing. Nobody independent has published results for it in at least two areas yet, so its weight is shared across the other four and the total is capped at 61 until they do or our own runs start."
          },
          {
            "key": "reach",
            "name": "Reach",
            "weight": 25,
            "effectiveWeight": 38.46,
            "score": 55,
            "points": 21.15,
            "reason": "Memory 30, the publisher's Kronos-mini with its tokeniser is 0.03 GB and even Kronos-base is 0.43 GB in FP32. Runtimes 0, it runs only through the publisher's own PyTorch package from GitHub, and none of the six named runtimes supports it. Builds 0, FP32 safetensors only. Terms 15, MIT with nothing added. Download 10, nothing is gated. Hosted 0, no inference provider or publisher API serves it.",
            "checked": "2026-10-25"
          },
          {
            "key": "risk",
            "name": "Risk",
            "weight": 20,
            "effectiveWeight": 30.77,
            "score": 50,
            "points": 15.38,
            "reason": "File safety 15, safetensors only, but the only documented loading path is the publisher's Python package in place of a runtime. Provenance 15, the files are in the NeoQuasar account the publisher's README links, with SHA-256 hashes. Safety evaluation 0, none published. Agent guidance 5, the README's general cautions that the fine-tuning pipeline isn't a production trading system. Licence risk 15, plain MIT; the broken licence link is charged to file health. Security programme 0, no reporting route found.",
            "checked": "2026-10-25"
          },
          {
            "key": "interface",
            "name": "Agent interface",
            "weight": 10,
            "effectiveWeight": 15.38,
            "score": 60,
            "points": 9.23,
            "reason": "Read for a forecasting model, the tool-calling, trained tool calling and structured-output lines stand for what it takes and returns. The card documents the DataFrame and timestamps it takes and the DataFrame it returns, and the publisher's KronosPredictor builds and parses them (25). It was trained on six K-line fields with calendar fields, one series per forecast (10). Output is a typed DataFrame indexed by the requested timestamps, the mean of the sampled paths (15). The input format is given only as code (10). Reasoning control and context score 0, with no reasoning and a 512-bar context (2,048 for Kronos-mini).",
            "checked": "2026-10-25"
          },
          {
            "key": "upkeep",
            "name": "Documentation and upkeep",
            "weight": 10,
            "effectiveWeight": 15.38,
            "score": 61,
            "points": 9.38,
            "reason": "Model card 11, intended use, architecture, parameters and context are there, limitations, languages and a cut-off aren't. Training data 20, the paper lists every exchange with asset and observation counts, the filtering and a June 2024 end. Last change 5, the reference code on 13 April 2026 and the files in mid-2025. Responsiveness 20, the publisher answered 10 of 18 Hugging Face discussions, though none since November 2025. Version policy 5, a first release. File health 0, the cards point to example data the repository has deleted and to licence and requirements links on a branch that doesn't exist.",
            "checked": "2026-10-25"
          }
        ],
        "assessment": {
          "date": "2026-10-25",
          "basis": "public evidence",
          "confidence": "low",
          "notes": {
            "capability": "Pending for this listing. We found no independent benchmark results for it in enough areas, so its weight is shared across the other four categories and its total is capped at 61.",
            "interface": "Read for a forecasting model, the tool-calling, trained tool calling and structured-output lines stand for what it takes and returns. The card documents the DataFrame and timestamps it takes and the DataFrame it returns, and the publisher's KronosPredictor builds and parses them (25). It was trained on six K-line fields with calendar fields, one series per forecast (10). Output is a typed DataFrame indexed by the requested timestamps, the mean of the sampled paths (15). The input format is given only as code (10). Reasoning control and context score 0, with no reasoning and a 512-bar context (2,048 for Kronos-mini).",
            "reach": "Memory 30, the publisher's Kronos-mini with its tokeniser is 0.03 GB and even Kronos-base is 0.43 GB in FP32. Runtimes 0, it runs only through the publisher's own PyTorch package from GitHub, and none of the six named runtimes supports it. Builds 0, FP32 safetensors only. Terms 15, MIT with nothing added. Download 10, nothing is gated. Hosted 0, no inference provider or publisher API serves it.",
            "risk": "File safety 15, safetensors only, but the only documented loading path is the publisher's Python package in place of a runtime. Provenance 15, the files are in the NeoQuasar account the publisher's README links, with SHA-256 hashes. Safety evaluation 0, none published. Agent guidance 5, the README's general cautions that the fine-tuning pipeline isn't a production trading system. Licence risk 15, plain MIT; the broken licence link is charged to file health. Security programme 0, no reporting route found.",
            "upkeep": "Model card 11, intended use, architecture, parameters and context are there, limitations, languages and a cut-off aren't. Training data 20, the paper lists every exchange with asset and observation counts, the filtering and a June 2024 end. Last change 5, the reference code on 13 April 2026 and the files in mid-2025. Responsiveness 20, the publisher answered 10 of 18 Hugging Face discussions, though none since November 2025. Version policy 5, a first release. File health 0, the cards point to example data the repository has deleted and to licence and requirements links on a branch that doesn't exist."
          },
          "sources": [
            {
              "what": "model card, read 25 October 2026 (the five cards differ only in tags)",
              "url": "https://huggingface.co/NeoQuasar/Kronos-base",
              "seen": "2026-10-25"
            },
            {
              "what": "the publisher's repositories, dates, downloads, gating and licence tags",
              "url": "https://huggingface.co/api/models?author=NeoQuasar\u0026full=true",
              "seen": "2026-10-25"
            },
            {
              "what": "parameter counts by dtype (F32) and empty provider mapping, likewise for the other four",
              "url": "https://huggingface.co/api/models/NeoQuasar/Kronos-base?expand[]=safetensors\u0026expand[]=inferenceProviderMapping",
              "seen": "2026-10-25"
            },
            {
              "what": "file list and sizes, likewise for the other four repositories",
              "url": "https://huggingface.co/api/models/NeoQuasar/Kronos-mini/tree/main?recursive=true",
              "seen": "2026-10-25"
            },
            {
              "what": "last commit per file, weights unchanged since June and July 2025",
              "url": "https://huggingface.co/api/models/NeoQuasar/Kronos-base/tree/main?expand=true",
              "seen": "2026-10-25"
            },
            {
              "what": "commit history, likewise for the other four",
              "url": "https://huggingface.co/api/models/NeoQuasar/Kronos-small/commits/main",
              "seen": "2026-10-25"
            },
            {
              "what": "discussions on all five repositories, authors, dates and owner flags counted, titles only",
              "url": "https://huggingface.co/api/models/NeoQuasar/Kronos-base/discussions",
              "seen": "2026-10-25"
            },
            {
              "what": "NeoQuasar is a user account named ShiYu",
              "url": "https://huggingface.co/api/users/NeoQuasar/overview",
              "seen": "2026-10-25"
            },
            {
              "what": "repository page (stars, issues, no releases, no security policy) and a shallow clone into scratch for the code, README, LICENSE and history",
              "url": "https://github.com/shiyu-coder/Kronos",
              "seen": "2026-10-25"
            },
            {
              "what": "MIT licence (200 on master, 404 on main)",
              "url": "https://raw.githubusercontent.com/shiyu-coder/Kronos/master/LICENSE",
              "seen": "2026-10-25"
            },
            {
              "what": "paper, architecture, training data table, filtering, June 2024 data end, publisher's results",
              "url": "https://arxiv.org/html/2508.02739v1",
              "seen": "2026-10-25"
            },
            {
              "what": "vLLM registry, no Kronos",
              "url": "https://raw.githubusercontent.com/vllm-project/vllm/main/vllm/model_executor/models/registry.py",
              "seen": "2026-10-25"
            },
            {
              "what": "404, no Kronos model in Transformers",
              "url": "https://raw.githubusercontent.com/huggingface/transformers/main/src/transformers/models/kronos/__init__.py",
              "seen": "2026-10-25"
            },
            {
              "what": "404, no Ollama library entry; Ollama's search finds only unrelated models",
              "url": "https://ollama.com/library/kronos",
              "seen": "2026-10-25"
            },
            {
              "what": "404, no LM Studio catalogue page",
              "url": "https://lmstudio.ai/models/neoquasar/kronos-base",
              "seen": "2026-10-25"
            },
            {
              "what": "live demo, Kronos-mini on BTC/USDT, last updated 4 July 2026",
              "url": "https://shiyu-coder.github.io/Kronos-demo/",
              "seen": "2026-10-25"
            }
          ],
          "openQuestions": [
            "Capability is pending. The checklist has no forecasting or time-series area, so a forecasting benchmark couldn't count however it was run, and four searches on 25 October 2026 found no independent benchmark of Kronos with a published method (two 2026 papers benchmarking time-series models on financial data don't include it); the total is capped at 61",
            "Unchecked, GitHub issues on shiyu-coder/Kronos (213 open), since the GitHub API isn't enabled for this session and the issues page was refused; responsiveness rests on the 18 Hugging Face discussions",
            "Unchecked, github.com/robots.txt, which our shell proxy and WebFetch both refused; the repository page was read once by WebFetch and the code by git clone",
            "The cards' licence link is read as broken from raw.githubusercontent.com (404 on main, 200 on master); github.com's own blob page answered 403, so a redirect there wasn't ruled out",
            "File health reads the deleted sample data as the files contradicting the card; a gentler reading of it as one disagreement gives 5 and Upkeep 66",
            "Whether the 13 April 2026 merges by the publisher of contributors' pull requests count as the publisher's own change; they're read as such, since the publisher merged them into master"
          ]
        },
        "negative": 0,
        "verdict": "Forecasts the next price bars from a history of up to 512 candlestick bars. Every size fits in under 0.5 GB and runs on a CPU, but only through the publisher's own Python package. Nobody has benchmarked it independently, its quickstart data was deleted from the repository, and the publisher hasn't replied since November 2025.",
        "bestFor": "Agents that need a quick local forecast or synthetic sample of a candlestick series on modest hardware, for research or as one input among others.",
        "strengths": [
          "MIT with nothing added, ungated, safetensors only with SHA-256 hashes",
          "Under 0.5 GB for every public size, so it runs on a CPU, Apple MPS or a small GPU",
          "Input and output documented as plain DataFrames, with batch prediction over many series",
          "Training data listed by exchange with observation counts, filtering and a June 2024 end"
        ],
        "weaknesses": [
          "Runs only through the publisher's own PyTorch code from GitHub, in no named runtime and with no quantised build",
          "A 512-bar context for the small and base sizes, and the mean forecast only, without quantiles",
          "The sample data the quickstart reads was deleted on 13 April 2026, and the cards' licence link points at a missing branch",
          "No safety results, no security contact and no publisher reply in its Hugging Face discussions since November 2025",
          "The publisher's forecasting results come from its own paper, with no independent benchmark found"
        ],
        "agentNotes": [
          "Clone https://github.com/shiyu-coder/Kronos and `pip install -r requirements.txt`, then load `Kronos.from_pretrained(\"NeoQuasar/Kronos-small\")` and `KronosTokenizer.from_pretrained(\"NeoQuasar/Kronos-Tokenizer-base\")` and wrap them in KronosPredictor",
          "Pair Kronos-mini with Kronos-Tokenizer-2k and Kronos-small or Kronos-base with Kronos-Tokenizer-base, the cards' pairing",
          "The examples read examples/data/XSHG_5min_600977.csv, which no longer exists; point them at tests/data/regression_input.csv, 2,500 five-minute bars",
          "Keep the lookback at or under max_context, 512 bars, and pass the future timestamps you want as y_timestamp, since the calendar fields come from them",
          "Raise sample_count for a steadier forecast, but expect only the mean back; call the lower-level auto_regressive_inference yourself if you need the paths",
          "Treat forecasts as model output to check, not signals; the README says its own backtest pipeline isn't a production trading system"
        ],
        "metrics": {
          "kind": "local",
          "measured": false
        },
        "reviewCount": 0,
        "avgRating": 0,
        "history": [
          {
            "basis": "public evidence",
            "confidence": "low",
            "grade": "C",
            "methodology": "0.6",
            "pending": [
              "performance",
              "tasks"
            ],
            "run": "2026-10-01",
            "runLabel": "October 2026 research run",
            "score": 55.2
          }
        ],
        "editorialScores": {
          "interface": 60,
          "reach": 55,
          "risk": 50,
          "upkeep": 61
        },
        "provenanceScore": 40,
        "checked": {
          "categories": {
            "interface": "2026-10-25",
            "reach": "2026-10-25",
            "risk": "2026-10-25",
            "upkeep": "2026-10-25"
          },
          "last": "2026-10-25",
          "oldest": "2026-10-25",
          "stale": false,
          "staleAfterDays": 90
        },
        "scoreHistory": [
          {
            "date": "2026-10-25",
            "grade": "C",
            "kind": "listed",
            "method": "0.6",
            "score": 55.2
          }
        ]
      },
      "connect": {
        "install": "git clone https://github.com/shiyu-coder/Kronos \u0026\u0026 cd Kronos\npip install -r requirements.txt   # Python 3.10+, PyTorch 2.0+\npython - \u003c\u003c'EOF'\nimport pandas as pd\nfrom model import Kronos, KronosTokenizer, KronosPredictor\ntok = KronosTokenizer.from_pretrained(\"NeoQuasar/Kronos-Tokenizer-base\")\nmodel = Kronos.from_pretrained(\"NeoQuasar/Kronos-small\")\np = KronosPredictor(model, tok, max_context=512)   # picks cuda, mps or cpu\ndf = pd.read_csv(\"tests/data/regression_input.csv\", parse_dates=[\"timestamps\"])\nx, y = df.iloc[:400], df[\"timestamps\"].iloc[400:520]\ncols = [\"open\", \"high\", \"low\", \"close\", \"volume\", \"amount\"]\nprint(p.predict(df=x[cols], x_timestamp=x[\"timestamps\"], y_timestamp=y, pred_len=120, T=1.0, top_p=0.9, sample_count=1).head())\nEOF"
      },
      "letme": {
        "capability": "https://letme.dev/model.open-weights",
        "tool": "https://letme.dev/kronos"
      },
      "sameCompany": [
        "github-mcp-server",
        "github-copilot-cli",
        "ai-job-search",
        "talorys"
      ],
      "weights": {
        "trainedBy": "Yu Shi and co-authors, Tsinghua University",
        "params": "4.1M to 102.3M",
        "context": 512,
        "goodFor": "Forecasting the next bars of a financial candlestick series, or sampling synthetic ones, on any machine, from a short history of open, high, low, close, volume and amount.",
        "runtimes": [],
        "builds": [
          {
            "name": "Kronos-mini",
            "url": "https://huggingface.co/NeoQuasar/Kronos-mini",
            "who": "publisher",
            "precision": "fp32",
            "sizeGB": 0.03,
            "memoryGB": 8,
            "format": "safetensors",
            "runsOn": [
              "PyTorch"
            ],
            "note": "32,304,418 bytes with Kronos-Tokenizer-2k, which it needs (16,451,361 for the model repository and 15,853,057 for the tokeniser). 2,048-bar context per the card"
          },
          {
            "name": "Kronos-small",
            "url": "https://huggingface.co/NeoQuasar/Kronos-small",
            "who": "publisher",
            "precision": "fp32",
            "sizeGB": 0.11,
            "memoryGB": 8,
            "format": "safetensors",
            "runsOn": [
              "PyTorch"
            ],
            "note": "114,844,309 bytes with Kronos-Tokenizer-base (98,991,258 and 15,853,051). The model the card's and README's examples load"
          },
          {
            "name": "Kronos-base",
            "url": "https://huggingface.co/NeoQuasar/Kronos-base",
            "who": "publisher",
            "precision": "fp32",
            "sizeGB": 0.43,
            "memoryGB": 8,
            "format": "safetensors",
            "runsOn": [
              "PyTorch"
            ],
            "note": "425,127,647 bytes with Kronos-Tokenizer-base (409,274,596 and 15,853,051). The largest public size"
          }
        ],
        "run": "git clone https://github.com/shiyu-coder/Kronos \u0026\u0026 cd Kronos \u0026\u0026 pip install -r requirements.txt, then in Python: KronosPredictor(Kronos.from_pretrained(\"NeoQuasar/Kronos-small\"), KronosTokenizer.from_pretrained(\"NeoQuasar/Kronos-Tokenizer-base\"), max_context=512).predict(df=..., x_timestamp=..., y_timestamp=..., pred_len=120)",
        "runBuild": "Kronos-small",
        "checked": "2026-10-25",
        "minMemoryGB": 8
      },
      "notable": [
        "The weight files were added on 30 June 2025 (Kronos-small, Kronos-base, Kronos-Tokenizer-base) and 1 July 2025 (Kronos-mini, Kronos-Tokenizer-2k) and haven't changed since. The last change to any of the five repositories was a card edit on 9 September 2025 (https://huggingface.co/api/models/NeoQuasar/Kronos-base/tree/main?expand=true)",
        "On 25 October 2026 GitHub showed 43.0k stars, 7.1k forks, 213 open issues, 72 open pull requests and no releases. The Hub counted 1,407,245 downloads in 30 days for Kronos-small, 719,333 for Kronos-base, 401,747 for Kronos-mini and 2,601,760 for Kronos-Tokenizer-base (https://github.com/shiyu-coder/Kronos)",
        "The paper says Kronos raises price-series forecasting RankIC by 93% over the leading time-series foundation model and 87% over the best non-pre-trained baseline, with 9% lower MAE in volatility forecasting and 22% better generative fidelity. These are the publisher's numbers, not our measurement (https://arxiv.org/abs/2508.02739)",
        "The README says the paper was accepted by AAAI 2026, and the paper's affiliations are Tsinghua University's Institute for Interdisciplinary Information Sciences and Department of Automation (https://arxiv.org/html/2508.02739v1)",
        "Kronos-large (499.2M parameters) is listed in the model zoo as not publicly available (https://huggingface.co/NeoQuasar/Kronos-base)",
        "The README's fine-tuning section says the pipeline is a demonstration, not a production-ready quantitative trading system, and that its signals are raw predictions (https://github.com/shiyu-coder/Kronos)",
        "The sample data the card, the README and three example scripts read (examples/data/XSHG_5min_600977.csv) was deleted from the repository by a contributor's pull request merged on 13 April 2026. tests/data/regression_input.csv still holds 2,500 five-minute bars (https://github.com/shiyu-coder/Kronos)",
        "The live demo forecasts BTC/USDT with Kronos-mini from 360 hours of Binance 1-hour bars and 30 sampled paths. On 25 October 2026 it showed a last update of 4 July 2026 (https://shiyu-coder.github.io/Kronos-demo/)",
        "Copies by others exist, MLX conversions from gxcsoccer and mirrors under several accounts. None of them is the publisher's (https://huggingface.co/api/models?search=kronos)"
      ],
      "area": "models",
      "details": [
        {
          "label": "Surface graded",
          "value": "The open weights in the five NeoQuasar repositories on Hugging Face (Kronos-mini, Kronos-small, Kronos-base and the two tokenisers), read from the cards, the files and their metadata, the publisher's GitHub repository (cloned into scratch), the paper and the runtimes' own pages on 25 October 2026. Nothing was downloaded or run. Nobody hosts an API for it"
        },
        {
          "label": "Model",
          "value": "Two stages. A tokeniser (3.96M parameters, a Transformer encoder and decoder) quantises each bar of open, high, low, close, volume and amount into a hierarchical pair of 10-bit subtokens, and a decoder-only Transformer predicts the next pair autoregressively, with learned embeddings for minute, hour, weekday, day and month. Kronos-mini 4,108,192 parameters (4 layers, d_model 256), Kronos-small 24,741,696 (8 layers, 512), Kronos-base 102,311,008 (12 layers, 832), all FP32, counted from the files. Trained from scratch, not a language model"
        },
        {
          "label": "Files",
          "value": "One model.safetensors and a config.json in each repository, with no pickle files and no Python. Kronos-mini 16.4 MB, Kronos-small 99.0 MB, Kronos-base 409.3 MB, Kronos-Tokenizer-base 15.8 MB, Kronos-Tokenizer-2k 15.8 MB. Hugging Face lists a SHA-256 for every weight file. The model code (model/kronos.py and model/module.py) is only in the GitHub repository and loads the files with huggingface_hub's PyTorchModelHubMixin"
        },
        {
          "label": "Licence",
          "value": "MIT in each card's metadata and License section, and MIT in the GitHub repository's LICENSE (copyright 2025 ShiYu). No LICENSE file in the Hugging Face repositories. No use policy. The cards link the licence at blob/main/LICENSE, but the repository's only branch is master, and raw.githubusercontent.com answers 404 for main and 200 for master"
        },
        {
          "label": "Runtimes",
          "value": "Only the publisher's own PyTorch code, cloned from GitHub, with torch 2.0 or later, huggingface_hub 0.33.1, safetensors, einops, pandas and numpy. Not in Transformers, vLLM, SGLang, llama.cpp, Ollama or LM Studio. Runs on CUDA, Apple MPS or CPU (the predictor picks one, and the regression test runs on CPU)"
        },
        {
          "label": "Input and output",
          "value": "predict takes a pandas DataFrame with open, high, low and close (volume and amount optional, filled with zeros if missing), the timestamps of those rows, the timestamps to forecast and pred_len. It returns a DataFrame of the six fields indexed by the requested timestamps. sample_count draws several paths and returns their mean, not the paths. predict_batch runs several series of equal length in parallel. Inputs are normalised per series and clipped at 5 standard deviations"
        },
        {
          "label": "Context",
          "value": "512 bars for Kronos-small and Kronos-base and 2,048 for Kronos-mini, per the cards. The predictor's max_context defaults to 512 and truncates longer histories. The paper says the context was limited to 512 tokens"
        },
        {
          "label": "Recommended settings",
          "value": "T 1.0, top_p 0.9 and sample_count 1 in the card's example, a lookback of 400 bars and no more than max_context. Pair Kronos-mini with Kronos-Tokenizer-2k and the other two with Kronos-Tokenizer-base"
        },
        {
          "label": "Hardware",
          "value": "No inference memory is stated. The paper trained and evaluated on 24 NVIDIA GeForce RTX 4090D GPUs. The largest public build with its tokeniser is 0.43 GB of FP32 files"
        },
        {
          "label": "Training data",
          "value": "Over 12 billion K-line records from 45 exchanges at seven frequencies from 1-minute to weekly, per the paper's abstract. Its appendix lists each exchange or index market with asset types, timeframes, asset and observation counts and start dates (mostly January 2020), describes the cleaning and filtering pipeline with its thresholds, and says crypto, futures and forex were up-weighted. The pre-training data extends to June 2024. The data itself isn't published and the vendor it came from isn't named. The appendix's text says over 40 exchanges in more than 30 countries"
        }
      ],
      "provenance": {
        "legalEntity": "No company. Yu Shi publishes it from the GitHub account shiyu-coder and the Hugging Face user account NeoQuasar (named ShiYu), with co-authors at Tsinghua University per the paper",
        "domain": "github.com",
        "domainRegistered": "",
        "endpointOnVendorDomain": null,
        "terms": "",
        "privacy": "",
        "statusPage": "",
        "changelog": "",
        "securityTxt": "none",
        "checked": "2026-10-25",
        "notes": [
          "NeoQuasar is a Hugging Face user account (fullname ShiYu), not an organisation. The GitHub README and the paper link it, so it's the publisher's own account.",
          "Nothing is hosted, so the MIT licence stands in for terms of service and no privacy policy is scored.",
          "No SECURITY.md, security contact or disclosure policy was found in the GitHub repository or on the cards.",
          "No changelog and no GitHub releases. The README's News section has three dated items, the last on 10 November 2025."
        ],
        "score": 40,
        "checks": [
          {
            "check": "Legal entity named",
            "value": "No company. Yu Shi publishes it from the GitHub account shiyu-coder and the Hugging Face user account NeoQuasar (named ShiYu), with co-authors at Tsinghua University per the paper",
            "points": 20,
            "max": 20,
            "state": "ok"
          },
          {
            "check": "Domain age",
            "value": "github.com, no registry record we could read",
            "points": 0,
            "max": 15,
            "state": "no"
          },
          {
            "check": "Endpoint on the vendor's domain",
            "value": "no hosted endpoint",
            "points": 0,
            "max": 0,
            "state": "na"
          },
          {
            "check": "Terms of service",
            "value": "nothing hosted, so the MIT licence stands in",
            "points": 10,
            "max": 10,
            "state": "ok"
          },
          {
            "check": "Privacy policy",
            "value": "nothing hosted, not scored",
            "points": 0,
            "max": 0,
            "state": "na"
          },
          {
            "check": "Status page",
            "value": "not found",
            "points": 0,
            "max": 10,
            "state": "no"
          },
          {
            "check": "Changelog",
            "value": "not found",
            "points": 0,
            "max": 10,
            "state": "no"
          },
          {
            "check": "security.txt",
            "value": "not found",
            "points": 0,
            "max": 10,
            "state": "no"
          }
        ]
      },
      "pageJsonUrl": "https://www.anchorterminal.com/tools/kronos.json",
      "live": {
        "slug": "kronos",
        "securityTxt": {
          "url": "https://github.com/.well-known/security.txt",
          "state": "valid",
          "expires": "2026-11-24T15:28:23z",
          "checkedAt": "2026-10-25T15:28:23.736926714Z"
        },
        "updatedAt": "2026-10-25T15:28:23.736926714Z"
      }
    },
    "verify": {
      "accepts": "a page under github.com/shiyu-coder, or the README of github.com/shiyu-coder/Kronos",
      "badgeUrl": "https://www.anchorterminal.com/badges/kronos.svg",
      "body": {
        "slug": "kronos",
        "url": "the page with the badge or the link"
      },
      "docs": "https://www.anchorterminal.com/builders/#verify",
      "effect": "none, it never changes a grade, rank or review",
      "endpoint": "https://www.anchorterminal.com/api/v1/verify",
      "listingUrl": "https://www.anchorterminal.com/tools/kronos",
      "mcpTool": "verify_listing",
      "recheck": "weekly; two failed checks in a row and it lapses, a later pass restores it",
      "snippets": {
        "html": "\u003ca href=\"https://www.anchorterminal.com/tools/kronos\"\u003e\u003cimg src=\"https://www.anchorterminal.com/badges/kronos.svg\" alt=\"Kronos on Anchor Terminal\" height=\"20\"\u003e\u003c/a\u003e",
        "markdown": "[![Kronos on Anchor Terminal](https://www.anchorterminal.com/badges/kronos.svg)](https://www.anchorterminal.com/tools/kronos)",
        "link": "\u003ca href=\"https://www.anchorterminal.com/tools/kronos\"\u003eKronos on Anchor Terminal\u003c/a\u003e"
      }
    }
  },
  "kind": "anchor.page",
  "links": {
    "api": "https://www.anchorterminal.com/api/v1/index.json",
    "html": "https://www.anchorterminal.com/tools/kronos",
    "json": "https://www.anchorterminal.com/tools/kronos.json",
    "llms": "https://www.anchorterminal.com/llms.txt",
    "markdown": "https://www.anchorterminal.com/tools/kronos.md",
    "mcp": "https://www.anchorterminal.com/mcp",
    "pricing": "https://www.anchorterminal.com/tools/kronos#pricing",
    "reviews": "https://www.anchorterminal.com/tools/kronos#reviews",
    "slim": "https://www.anchorterminal.com/tools/kronos.min.md"
  },
  "markdown": "## Overview\n\n**Grade C · 55.2/100 · rank #86 of 110 open-weight models · #68 in Open-weight models · not agent-ready · confidence low**\n\n\nMore from Yu Shi, listed separately because each is its own product: [GitHub MCP Server](https://www.anchorterminal.com/tools/github-mcp-server.md) (Code \u0026 developer platforms), [GitHub Copilot CLI](https://www.anchorterminal.com/tools/github-copilot-cli.md) (Agent harnesses), [AI Job Search](https://www.anchorterminal.com/tools/ai-job-search.md) (Personal assistants), [Talorys](https://www.anchorterminal.com/tools/talorys.md) (Personal assistants).\n\n## Assessment\n\nForecasts the next price bars from a history of up to 512 candlestick bars. Every size fits in under 0.5 GB and runs on a CPU, but only through the publisher's own Python package. Nobody has benchmarked it independently, its quickstart data was deleted from the repository, and the publisher hasn't replied since November 2025.\n\n## Facts\n\n| Field | Value |\n| --- | --- |\n| Vendor | Yu Shi (shiyu-coder, NeoQuasar), Tsinghua University (https://github.com/shiyu-coder/Kronos) |\n| Kind | Open-weight model |\n| Category | Open-weight models (https://www.anchorterminal.com/categories/open-weights) |\n| Auth | None · The weights download from Hugging Face with no account, token or terms to accept. The Hub reports gated false for all five NeoQuasar repositories, and the code is in a public GitHub repository (checked 2026-10-25). |\n| Pricing | Free (Free) · Free to download and use, commercial use included, under the MIT licence named in each card's metadata and in the GitHub repository's LICENSE. Nobody sells per-call access and no inference provider serves it. Running it costs the hardware, 0.03 GB for Kronos-mini with its tokeniser and 0.43 GB for Kronos-base with its tokeniser, in FP32 (checked 2026-10-25). |\n| x402 | No · Weights you download and run yourself, so there is nothing to pay per call and no payment protocol (checked 2026-10-25). |\n| Licence | MIT |\n| Packages | huggingface: `NeoQuasar/Kronos-mini`; huggingface: `NeoQuasar/Kronos-small`; huggingface: `NeoQuasar/Kronos-base`; huggingface: `NeoQuasar/Kronos-Tokenizer-base`; huggingface: `NeoQuasar/Kronos-Tokenizer-2k` |\n| Source | https://github.com/shiyu-coder/Kronos |\n| Docs | https://github.com/shiyu-coder/Kronos |\n| llms.txt | not found |\n| Last release | 2026-04-13 |\n| GitHub stars | 43,000 (as of 2026-10-25) |\n| Surface graded | The open weights in the five NeoQuasar repositories on Hugging Face (Kronos-mini, Kronos-small, Kronos-base and the two tokenisers), read from the cards, the files and their metadata, the publisher's GitHub repository (cloned into scratch), the paper and the runtimes' own pages on 25 October 2026. Nothing was downloaded or run. Nobody hosts an API for it |\n| Model | Two stages. A tokeniser (3.96M parameters, a Transformer encoder and decoder) quantises each bar of open, high, low, close, volume and amount into a hierarchical pair of 10-bit subtokens, and a decoder-only Transformer predicts the next pair autoregressively, with learned embeddings for minute, hour, weekday, day and month. Kronos-mini 4,108,192 parameters (4 layers, d_model 256), Kronos-small 24,741,696 (8 layers, 512), Kronos-base 102,311,008 (12 layers, 832), all FP32, counted from the files. Trained from scratch, not a language model |\n| Files | One model.safetensors and a config.json in each repository, with no pickle files and no Python. Kronos-mini 16.4 MB, Kronos-small 99.0 MB, Kronos-base 409.3 MB, Kronos-Tokenizer-base 15.8 MB, Kronos-Tokenizer-2k 15.8 MB. Hugging Face lists a SHA-256 for every weight file. The model code (model/kronos.py and model/module.py) is only in the GitHub repository and loads the files with huggingface_hub's PyTorchModelHubMixin |\n| Licence | MIT in each card's metadata and License section, and MIT in the GitHub repository's LICENSE (copyright 2025 ShiYu). No LICENSE file in the Hugging Face repositories. No use policy. The cards link the licence at blob/main/LICENSE, but the repository's only branch is master, and raw.githubusercontent.com answers 404 for main and 200 for master |\n| Runtimes | Only the publisher's own PyTorch code, cloned from GitHub, with torch 2.0 or later, huggingface_hub 0.33.1, safetensors, einops, pandas and numpy. Not in Transformers, vLLM, SGLang, llama.cpp, Ollama or LM Studio. Runs on CUDA, Apple MPS or CPU (the predictor picks one, and the regression test runs on CPU) |\n| Input and output | predict takes a pandas DataFrame with open, high, low and close (volume and amount optional, filled with zeros if missing), the timestamps of those rows, the timestamps to forecast and pred_len. It returns a DataFrame of the six fields indexed by the requested timestamps. sample_count draws several paths and returns their mean, not the paths. predict_batch runs several series of equal length in parallel. Inputs are normalised per series and clipped at 5 standard deviations |\n| Context | 512 bars for Kronos-small and Kronos-base and 2,048 for Kronos-mini, per the cards. The predictor's max_context defaults to 512 and truncates longer histories. The paper says the context was limited to 512 tokens |\n| Recommended settings | T 1.0, top_p 0.9 and sample_count 1 in the card's example, a lookback of 400 bars and no more than max_context. Pair Kronos-mini with Kronos-Tokenizer-2k and the other two with Kronos-Tokenizer-base |\n| Hardware | No inference memory is stated. The paper trained and evaluated on 24 NVIDIA GeForce RTX 4090D GPUs. The largest public build with its tokeniser is 0.43 GB of FP32 files |\n| Training data | Over 12 billion K-line records from 45 exchanges at seven frequencies from 1-minute to weekly, per the paper's abstract. Its appendix lists each exchange or index market with asset types, timeframes, asset and observation counts and start dates (mostly January 2020), describes the cleaning and filtering pipeline with its thresholds, and says crypto, futures and forex were up-weighted. The pre-training data extends to June 2024. The data itself isn't published and the vendor it came from isn't named. The appendix's text says over 40 exchanges in more than 30 countries |\n| Capabilities | model.open-weights |\n| Tags | open-weights, model, time-series, forecasting, finance, candlestick, k-line, mit, safetensors, self-hosted |\n| JSON | https://www.anchorterminal.com/api/v1/tools/kronos.json |\n\n## Score breakdown (methodology v0.6, October 2026 research run)\n\nAssessed 2026-10-25 from public evidence against the published checklist (https://www.anchorterminal.com/benchmark/#checklist). Confidence: low. Performance and Task success pending (no score, not in the total); the total is Σ(score × weight) ÷ 80 over the 7 assessed categories. \"This run\" is each category's share of the 100 points.\n\n| Category | Weight | This run | Score (0–100) | Points |\n| --- | --- | --- | --- | --- |\n| Capability | 35% | pending | pending | n/a |\n| Reach | 25% | 38.5 | 55 | 21.1 |\n| Risk | 20% | 30.8 | 50 | 15.4 |\n| Agent interface | 10% | 15.4 | 60 | 9.2 |\n| Documentation and upkeep | 10% | 15.4 | 61 | 9.4 |\n| Negative events | up to −15 | up to −15 | none recorded | 0 |\n| **Total** | | | | **55.2 → C** |\n\n### Why each score\n\n- Capability: Pending for this listing. Nobody independent has published results for it in at least two areas yet, so its weight is shared across the other four and the total is capped at 61 until they do or our own runs start.\n- Reach 55 (checked 2026-10-25): Memory 30, the publisher's Kronos-mini with its tokeniser is 0.03 GB and even Kronos-base is 0.43 GB in FP32. Runtimes 0, it runs only through the publisher's own PyTorch package from GitHub, and none of the six named runtimes supports it. Builds 0, FP32 safetensors only. Terms 15, MIT with nothing added. Download 10, nothing is gated. Hosted 0, no inference provider or publisher API serves it.\n- Risk 50 (checked 2026-10-25): File safety 15, safetensors only, but the only documented loading path is the publisher's Python package in place of a runtime. Provenance 15, the files are in the NeoQuasar account the publisher's README links, with SHA-256 hashes. Safety evaluation 0, none published. Agent guidance 5, the README's general cautions that the fine-tuning pipeline isn't a production trading system. Licence risk 15, plain MIT; the broken licence link is charged to file health. Security programme 0, no reporting route found.\n- Agent interface 60 (checked 2026-10-25): Read for a forecasting model, the tool-calling, trained tool calling and structured-output lines stand for what it takes and returns. The card documents the DataFrame and timestamps it takes and the DataFrame it returns, and the publisher's KronosPredictor builds and parses them (25). It was trained on six K-line fields with calendar fields, one series per forecast (10). Output is a typed DataFrame indexed by the requested timestamps, the mean of the sampled paths (15). The input format is given only as code (10). Reasoning control and context score 0, with no reasoning and a 512-bar context (2,048 for Kronos-mini).\n- Documentation and upkeep 61 (checked 2026-10-25): Model card 11, intended use, architecture, parameters and context are there, limitations, languages and a cut-off aren't. Training data 20, the paper lists every exchange with asset and observation counts, the filtering and a June 2024 end. Last change 5, the reference code on 13 April 2026 and the files in mid-2025. Responsiveness 20, the publisher answered 10 of 18 Hugging Face discussions, though none since November 2025. Version policy 5, a first release. File health 0, the cards point to example data the repository has deleted and to licence and requirements links on a branch that doesn't exist.\n\nFix list for a coding agent, everything this grade says the listing lacks, the biggest gain first (14 items): https://www.anchorterminal.com/fixes/kronos.md (JSON https://www.anchorterminal.com/fixes/kronos.json)\n\n### What we couldn't check\n\n- Capability is pending. The checklist has no forecasting or time-series area, so a forecasting benchmark couldn't count however it was run, and four searches on 25 October 2026 found no independent benchmark of Kronos with a published method (two 2026 papers benchmarking time-series models on financial data don't include it); the total is capped at 61\n- Unchecked, GitHub issues on shiyu-coder/Kronos (213 open), since the GitHub API isn't enabled for this session and the issues page was refused; responsiveness rests on the 18 Hugging Face discussions\n- Unchecked, github.com/robots.txt, which our shell proxy and WebFetch both refused; the repository page was read once by WebFetch and the code by git clone\n- The cards' licence link is read as broken from raw.githubusercontent.com (404 on main, 200 on master); github.com's own blob page answered 403, so a redirect there wasn't ruled out\n- File health reads the deleted sample data as the files contradicting the card; a gentler reading of it as one disagreement gives 5 and Upkeep 66\n- Whether the 13 April 2026 merges by the publisher of contributors' pull requests count as the publisher's own change; they're read as such, since the publisher merged them into master\n\n### Sources\n\n- model card, read 25 October 2026 (the five cards differ only in tags): \u003chttps://huggingface.co/NeoQuasar/Kronos-base\u003e (seen 2026-10-25)\n- the publisher's repositories, dates, downloads, gating and licence tags: \u003chttps://huggingface.co/api/models?author=NeoQuasar\u0026full=true\u003e (seen 2026-10-25)\n- parameter counts by dtype (F32) and empty provider mapping, likewise for the other four: \u003chttps://huggingface.co/api/models/NeoQuasar/Kronos-base?expand[]=safetensors\u0026expand[]=inferenceProviderMapping\u003e (seen 2026-10-25)\n- file list and sizes, likewise for the other four repositories: \u003chttps://huggingface.co/api/models/NeoQuasar/Kronos-mini/tree/main?recursive=true\u003e (seen 2026-10-25)\n- last commit per file, weights unchanged since June and July 2025: \u003chttps://huggingface.co/api/models/NeoQuasar/Kronos-base/tree/main?expand=true\u003e (seen 2026-10-25)\n- commit history, likewise for the other four: \u003chttps://huggingface.co/api/models/NeoQuasar/Kronos-small/commits/main\u003e (seen 2026-10-25)\n- discussions on all five repositories, authors, dates and owner flags counted, titles only: \u003chttps://huggingface.co/api/models/NeoQuasar/Kronos-base/discussions\u003e (seen 2026-10-25)\n- NeoQuasar is a user account named ShiYu: \u003chttps://huggingface.co/api/users/NeoQuasar/overview\u003e (seen 2026-10-25)\n- repository page (stars, issues, no releases, no security policy) and a shallow clone into scratch for the code, README, LICENSE and history: \u003chttps://github.com/shiyu-coder/Kronos\u003e (seen 2026-10-25)\n- MIT licence (200 on master, 404 on main): \u003chttps://raw.githubusercontent.com/shiyu-coder/Kronos/master/LICENSE\u003e (seen 2026-10-25)\n- paper, architecture, training data table, filtering, June 2024 data end, publisher's results: \u003chttps://arxiv.org/html/2508.02739v1\u003e (seen 2026-10-25)\n- vLLM registry, no Kronos: \u003chttps://raw.githubusercontent.com/vllm-project/vllm/main/vllm/model_executor/models/registry.py\u003e (seen 2026-10-25)\n- 404, no Kronos model in Transformers: \u003chttps://raw.githubusercontent.com/huggingface/transformers/main/src/transformers/models/kronos/__init__.py\u003e (seen 2026-10-25)\n- 404, no Ollama library entry; Ollama's search finds only unrelated models: \u003chttps://ollama.com/library/kronos\u003e (seen 2026-10-25)\n- 404, no LM Studio catalogue page: \u003chttps://lmstudio.ai/models/neoquasar/kronos-base\u003e (seen 2026-10-25)\n- live demo, Kronos-mini on BTC/USDT, last updated 4 July 2026: \u003chttps://shiyu-coder.github.io/Kronos-demo/\u003e (seen 2026-10-25)\n\n## Score history\n\nEvery move of a score since the listing was graded, newest first, with its cause. A score can't move without an entry here. JSON: https://www.anchorterminal.com/api/v1/tools/kronos/history.json\n\n- 2026-10-25: graded 55.2 C (methodology 0.6)\n\n## Who's behind it (provenance 40/100, checked 2026-10-25)\n\n| Check | Finding | Points |\n| --- | --- | --- |\n| Legal entity named | No company. Yu Shi publishes it from the GitHub account shiyu-coder and the Hugging Face user account NeoQuasar (named ShiYu), with co-authors at Tsinghua University per the paper | 20/20 |\n| Domain age | github.com, no registry record we could read | 0/15 |\n| Endpoint on the vendor's domain | no hosted endpoint | n/a |\n| Terms of service | nothing hosted, so the MIT licence stands in | 10/10 |\n| Privacy policy | nothing hosted, not scored | n/a |\n| Status page | not found | 0/10 |\n| Changelog | not found | 0/10 |\n| security.txt | not found | 0/10 |\n\nNeoQuasar is a Hugging Face user account (fullname ShiYu), not an organisation. The GitHub README and the paper link it, so it's the publisher's own account.\n\nNothing is hosted, so the MIT licence stands in for terms of service and no privacy policy is scored.\n\nNo SECURITY.md, security contact or disclosure policy was found in the GitHub repository or on the cards.\n\nNo changelog and no GitHub releases. The README's News section has three dated items, the last on 10 November 2025.\n\n### Terms and privacy, as read\n\nA reading by a fixed set of rules, each answered with the vendor's own sentence. Not legal advice.\n\n**Terms of service**. Nothing is hosted by the vendor, so there are no terms of service to read. The MIT licence stands in and the check scores in full.\n\n\n**Privacy policy**. Nothing is hosted by the vendor, so there is no privacy policy to read and the check isn't scored.\n\n\n## Live (updated 2026-10-25 15:28 UTC)\n\n- security.txt: valid, expires 2026-11-24T15:28:23z\n- Always current: https://www.anchorterminal.com/api/v1/live/kronos.json\n\n## Probe metrics\n\nNot measured yet. Our benchmark probes haven't run, so there's no availability, latency or error rate from a run and Performance is pending. Live uptime, where we poll the endpoint, is under Live and doesn't change the score.\n\n## Strengths\n\n- MIT with nothing added, ungated, safetensors only with SHA-256 hashes\n- Under 0.5 GB for every public size, so it runs on a CPU, Apple MPS or a small GPU\n- Input and output documented as plain DataFrames, with batch prediction over many series\n- Training data listed by exchange with observation counts, filtering and a June 2024 end\n\n## Weaknesses\n\n- Runs only through the publisher's own PyTorch code from GitHub, in no named runtime and with no quantised build\n- A 512-bar context for the small and base sizes, and the mean forecast only, without quantiles\n- The sample data the quickstart reads was deleted on 13 April 2026, and the cards' licence link points at a missing branch\n- No safety results, no security contact and no publisher reply in its Hugging Face discussions since November 2025\n- The publisher's forecasting results come from its own paper, with no independent benchmark found\n\n\n\n## Before you call it (notes for agents)\n\n1. Clone https://github.com/shiyu-coder/Kronos and `pip install -r requirements.txt`, then load `Kronos.from_pretrained(\"NeoQuasar/Kronos-small\")` and `KronosTokenizer.from_pretrained(\"NeoQuasar/Kronos-Tokenizer-base\")` and wrap them in KronosPredictor\n2. Pair Kronos-mini with Kronos-Tokenizer-2k and Kronos-small or Kronos-base with Kronos-Tokenizer-base, the cards' pairing\n3. The examples read examples/data/XSHG_5min_600977.csv, which no longer exists; point them at tests/data/regression_input.csv, 2,500 five-minute bars\n4. Keep the lookback at or under max_context, 512 bars, and pass the future timestamps you want as y_timestamp, since the calendar fields come from them\n5. Raise sample_count for a steadier forecast, but expect only the mean back; call the lower-level auto_regressive_inference yourself if you need the paths\n6. Treat forecasts as model output to check, not signals; the README says its own backtest pipeline isn't a production trading system\n\n## Get started\n\nInstall:\n\n```bash\ngit clone https://github.com/shiyu-coder/Kronos \u0026\u0026 cd Kronos\npip install -r requirements.txt   # Python 3.10+, PyTorch 2.0+\npython - \u003c\u003c'EOF'\nimport pandas as pd\nfrom model import Kronos, KronosTokenizer, KronosPredictor\ntok = KronosTokenizer.from_pretrained(\"NeoQuasar/Kronos-Tokenizer-base\")\nmodel = Kronos.from_pretrained(\"NeoQuasar/Kronos-small\")\np = KronosPredictor(model, tok, max_context=512)   # picks cuda, mps or cpu\ndf = pd.read_csv(\"tests/data/regression_input.csv\", parse_dates=[\"timestamps\"])\nx, y = df.iloc[:400], df[\"timestamps\"].iloc[400:520]\ncols = [\"open\", \"high\", \"low\", \"close\", \"volume\", \"amount\"]\nprint(p.predict(df=x[cols], x_timestamp=x[\"timestamps\"], y_timestamp=y, pred_len=120, T=1.0, top_p=0.9, sample_count=1).head())\nEOF\n```\n\n## Similar tools\n\nRanked by shared capabilities, then score. Same-category tools with no shared capability key are listed last.\n\n| Tool | Grade | Score | Rank | Shared capabilities | x402 | Markdown |\n| --- | --- | --- | --- | --- | --- | --- |\n| Qwen3-Embedding-8B | BB | 74.7 | open-weight #1 | model.open-weights | no | https://www.anchorterminal.com/tools/qwen3-embedding-8b.md |\n| Gemma 4 31B IT | BB | 74.1 | open-weight #2 | model.open-weights | no | https://www.anchorterminal.com/tools/gemma-4-31b-it.md |\n| Qwen3-Embedding-4B | BB | 72.9 | open-weight #3 | model.open-weights | no | https://www.anchorterminal.com/tools/qwen3-embedding-4b.md |\n| Gemma 4 26B A4B IT | BB | 72.2 | open-weight #4 | model.open-weights | no | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.md |\n| Gemma 4 12B IT | BB | 71.3 | open-weight #5 | model.open-weights | no | https://www.anchorterminal.com/tools/gemma-4-12b-it.md |\n| gpt-oss-120b | BB | 70.2 | open-weight #6 | model.open-weights | no | https://www.anchorterminal.com/tools/gpt-oss-120b.md |\n\n## Panel reviews (0)\n\nReviewed by the Anchor panel (https://www.anchorterminal.com/reviewers/index.md): .\n\nDesk reviews, written from public documentation, pricing, terms, source and status history on 1 October 2026. No calls made. For a desk review, the outcome says whether the reviewer's questions could be answered from public material: success, partial or failure. How reviews work: https://www.anchorterminal.com/reviews/how-it-works.md\n\n## Notable\n\n- The weight files were added on 30 June 2025 (Kronos-small, Kronos-base, Kronos-Tokenizer-base) and 1 July 2025 (Kronos-mini, Kronos-Tokenizer-2k) and haven't changed since. The last change to any of the five repositories was a card edit on 9 September 2025 (source: \u003chttps://huggingface.co/api/models/NeoQuasar/Kronos-base/tree/main?expand=true\u003e)\n- On 25 October 2026 GitHub showed 43.0k stars, 7.1k forks, 213 open issues, 72 open pull requests and no releases. The Hub counted 1,407,245 downloads in 30 days for Kronos-small, 719,333 for Kronos-base, 401,747 for Kronos-mini and 2,601,760 for Kronos-Tokenizer-base (source: \u003chttps://github.com/shiyu-coder/Kronos\u003e)\n- The paper says Kronos raises price-series forecasting RankIC by 93% over the leading time-series foundation model and 87% over the best non-pre-trained baseline, with 9% lower MAE in volatility forecasting and 22% better generative fidelity. These are the publisher's numbers, not our measurement (source: \u003chttps://arxiv.org/abs/2508.02739\u003e)\n- The README says the paper was accepted by AAAI 2026, and the paper's affiliations are Tsinghua University's Institute for Interdisciplinary Information Sciences and Department of Automation (source: \u003chttps://arxiv.org/html/2508.02739v1\u003e)\n- Kronos-large (499.2M parameters) is listed in the model zoo as not publicly available (source: \u003chttps://huggingface.co/NeoQuasar/Kronos-base\u003e)\n- The README's fine-tuning section says the pipeline is a demonstration, not a production-ready quantitative trading system, and that its signals are raw predictions (source: \u003chttps://github.com/shiyu-coder/Kronos\u003e)\n- The sample data the card, the README and three example scripts read (examples/data/XSHG_5min_600977.csv) was deleted from the repository by a contributor's pull request merged on 13 April 2026. tests/data/regression_input.csv still holds 2,500 five-minute bars (source: \u003chttps://github.com/shiyu-coder/Kronos\u003e)\n- The live demo forecasts BTC/USDT with Kronos-mini from 360 hours of Binance 1-hour bars and 30 sampled paths. On 25 October 2026 it showed a last update of 4 July 2026 (source: \u003chttps://shiyu-coder.github.io/Kronos-demo/\u003e)\n- Copies by others exist, MLX conversions from gxcsoccer and mirrors under several accounts. None of them is the publisher's (source: \u003chttps://huggingface.co/api/models?search=kronos\u003e)\n\n- #68 of 81 in Best open-weight models you download and run yourself for AI agents: https://www.anchorterminal.com/best/open-weights/index.md\n- All 3260 open weights comparisons: https://www.anchorterminal.com/compare/open-weights/index.md\n\n## Compare\n\n- [Agents-A1-4B vs Kronos](https://www.anchorterminal.com/compare/agents-a1-4b-vs-kronos.md): C 61 vs C 55.2\n- [Apertus v1.5 8B vs Kronos](https://www.anchorterminal.com/compare/apertus-v1-5-8b-vs-kronos.md): C 61 vs C 55.2\n- [Apodex-1.1-mini vs Kronos](https://www.anchorterminal.com/compare/apodex-1-1-mini-vs-kronos.md): C 61 vs C 55.2\n- [BigBang-v1 vs Kronos](https://www.anchorterminal.com/compare/bigbang-v1-vs-kronos.md): C 54.4 vs C 55.2\n- [Carbon-X vs Kronos](https://www.anchorterminal.com/compare/carbon-x-vs-kronos.md): C 61 vs C 55.2\n- [Cosmos-Reason2-2B vs Kronos](https://www.anchorterminal.com/compare/cosmos-reason2-2b-vs-kronos.md): C 59.5 vs C 55.2\n- [Darwin-36B-Opus vs Kronos](https://www.anchorterminal.com/compare/darwin-36b-opus-vs-kronos.md): C 61 vs C 55.2\n- [DeepSeek-V3.2 vs Kronos](https://www.anchorterminal.com/compare/deepseek-v3-2-vs-kronos.md): C 58.6 vs C 55.2\n- [DeepSeek V4.1 Flash vs Kronos](https://www.anchorterminal.com/compare/deepseek-v4-1-flash-vs-kronos.md): C 59 vs C 55.2\n- [DeepSeek-V4-Flash-0731 vs Kronos](https://www.anchorterminal.com/compare/deepseek-v4-flash-0731-vs-kronos.md): B 62.4 vs C 55.2\n- [DiffusionGemma 26B A4B IT vs Kronos](https://www.anchorterminal.com/compare/diffusiongemma-26b-a4b-it-vs-kronos.md): D 50 vs C 55.2\n- [Gemma 4 12B IT vs Kronos](https://www.anchorterminal.com/compare/gemma-4-12b-it-vs-kronos.md): BB 71.3 vs C 55.2\n- [Gemma 4 26B A4B IT vs Kronos](https://www.anchorterminal.com/compare/gemma-4-26b-a4b-it-vs-kronos.md): BB 72.2 vs C 55.2\n- [Gemma 4 31B IT vs Kronos](https://www.anchorterminal.com/compare/gemma-4-31b-it-vs-kronos.md): BB 74.1 vs C 55.2\n- [Gemma 4 E2B IT vs Kronos](https://www.anchorterminal.com/compare/gemma-4-e2b-it-vs-kronos.md): C 60.4 vs C 55.2\n- [Gemma 4 E4B IT vs Kronos](https://www.anchorterminal.com/compare/gemma-4-e4b-it-vs-kronos.md): B 63.2 vs C 55.2\n- [GLM-4.5-Air vs Kronos](https://www.anchorterminal.com/compare/glm-4-5-air-vs-kronos.md): C 59.1 vs C 55.2\n- [GLM-4.7-Flash vs Kronos](https://www.anchorterminal.com/compare/glm-4-7-flash-vs-kronos.md): B 63 vs C 55.2\n- [GLM-5.3-Flash vs Kronos](https://www.anchorterminal.com/compare/glm-5-3-flash-vs-kronos.md): B 65.1 vs C 55.2\n- [GLM-5.3 vs Kronos](https://www.anchorterminal.com/compare/glm-5-3-vs-kronos.md): B 64.6 vs C 55.2\n- [gpt-oss-120b vs Kronos](https://www.anchorterminal.com/compare/gpt-oss-120b-vs-kronos.md): BB 70.2 vs C 55.2\n- [gpt-oss-20b vs Kronos](https://www.anchorterminal.com/compare/gpt-oss-20b-vs-kronos.md): B 69.9 vs C 55.2\n- [Granite 4.2 3B vs Kronos](https://www.anchorterminal.com/compare/granite-4-2-3b-vs-kronos.md): C 59.8 vs C 55.2\n- [Hy3 vs Kronos](https://www.anchorterminal.com/compare/hy3-vs-kronos.md): B 62.6 vs C 55.2\n- [Inkling-Small vs Kronos](https://www.anchorterminal.com/compare/inkling-small-vs-kronos.md): B 66.5 vs C 55.2\n- [Inkling vs Kronos](https://www.anchorterminal.com/compare/inkling-vs-kronos.md): B 64.2 vs C 55.2\n- [InternVL3.5-GPT-OSS-20B-A4B-Preview vs Kronos](https://www.anchorterminal.com/compare/internvl3-5-gpt-oss-20b-a4b-preview-vs-kronos.md): D 51.2 vs C 55.2\n- [Kimi K3 vs Kronos](https://www.anchorterminal.com/compare/kimi-k3-vs-kronos.md): B 62.4 vs C 55.2\n- [Kolibri-1 vs Kronos](https://www.anchorterminal.com/compare/kolibri-1-vs-kronos.md): C 61 vs C 55.2\n- [Kronos vs Laguna M.1](https://www.anchorterminal.com/compare/kronos-vs-laguna-m-1.md): C 55.2 vs C 55.5\n- [Kronos vs Laguna S 2.1](https://www.anchorterminal.com/compare/kronos-vs-laguna-s-2-1.md): C 55.2 vs C 61\n- [Kronos vs Laguna XS 2.1](https://www.anchorterminal.com/compare/kronos-vs-laguna-xs-2-1.md): C 55.2 vs C 61\n- [Kronos vs LFM2.5-2.6B](https://www.anchorterminal.com/compare/kronos-vs-lfm2-5-2-6b.md): C 55.2 vs D 51.5\n- [Kronos vs Ling-3.0-tiny](https://www.anchorterminal.com/compare/kronos-vs-ling-3-0-tiny.md): C 55.2 vs C 55\n- [Kronos vs MiMo-V2.6 Distill Qwen 9B](https://www.anchorterminal.com/compare/kronos-vs-mimo-v2-6-distill-qwen-9b.md): C 55.2 vs C 61\n- [Kronos vs MiniCPM-o 4.5](https://www.anchorterminal.com/compare/kronos-vs-minicpm-o-4-5.md): C 55.2 vs C 61\n- [Kronos vs MiniCPM5-2B](https://www.anchorterminal.com/compare/kronos-vs-minicpm5-2b.md): C 55.2 vs C 60.2\n- [Kronos vs MiniMax-M2.5](https://www.anchorterminal.com/compare/kronos-vs-minimax-m2-5.md): C 55.2 vs C 57.5\n- [Kronos vs MiniMax-M2.7](https://www.anchorterminal.com/compare/kronos-vs-minimax-m2-7.md): C 55.2 vs C 56.1\n- [Kronos vs MiniMax-M3.1-Flash](https://www.anchorterminal.com/compare/kronos-vs-minimax-m3-1-flash.md): C 55.2 vs D 50.5\n- [Kronos vs Muse Glimmer 30B](https://www.anchorterminal.com/compare/kronos-vs-muse-glimmer-30b.md): C 55.2 vs BB 70\n- [Kronos vs NVIDIA Nemotron 3.5 Bolt E4B 4GB](https://www.anchorterminal.com/compare/kronos-vs-nemotron-3-5-bolt-e4b-4gb.md): C 55.2 vs C 61\n- [Kronos vs Nemotron 3.5 Lightning 30B A3B](https://www.anchorterminal.com/compare/kronos-vs-nemotron-3-5-lightning-30b-a3b.md): C 55.2 vs B 64.2\n- [Kronos vs Nemotron 3 Nano 30B A3B](https://www.anchorterminal.com/compare/kronos-vs-nemotron-3-nano-30b-a3b.md): C 55.2 vs B 67.1\n- [Kronos vs Nemotron 3 Nano 4B](https://www.anchorterminal.com/compare/kronos-vs-nemotron-3-nano-4b.md): C 55.2 vs C 56.8\n- [Kronos vs Nemotron 3 Nano Omni 30B A3B Reasoning](https://www.anchorterminal.com/compare/kronos-vs-nemotron-3-nano-omni-30b-a3b-reasoning.md): C 55.2 vs C 60.7\n- [Kronos vs Nemotron 3 Super 120B A12B](https://www.anchorterminal.com/compare/kronos-vs-nemotron-3-super-120b-a12b.md): C 55.2 vs B 66.2\n- [Kronos vs Nemotron 3 Ultra 550B A55B](https://www.anchorterminal.com/compare/kronos-vs-nemotron-3-ultra-550b-a55b.md): C 55.2 vs B 68.8\n- [Kronos vs occamy-1.0](https://www.anchorterminal.com/compare/kronos-vs-occamy-1-0.md): C 55.2 vs C 61\n- [Kronos vs Olmo 3 7B Think](https://www.anchorterminal.com/compare/kronos-vs-olmo-3-7b-think.md): C 55.2 vs D 52.3\n- [Kronos vs Ornith-1.5-35B-A3B](https://www.anchorterminal.com/compare/kronos-vs-ornith-1-5-35b-a3b.md): C 55.2 vs C 61\n- [Kronos vs Ornith-1.5-397B](https://www.anchorterminal.com/compare/kronos-vs-ornith-1-5-397b.md): C 55.2 vs C 59\n- [Kronos vs Ornith-1.5-9B](https://www.anchorterminal.com/compare/kronos-vs-ornith-1-5-9b.md): C 55.2 vs C 61\n- [Kronos vs Qwen3-30B-A3B-Instruct-2507](https://www.anchorterminal.com/compare/kronos-vs-qwen3-30b-a3b-instruct-2507.md): C 55.2 vs C 56.6\n- [Kronos vs Qwen3-4B-Instruct-2507](https://www.anchorterminal.com/compare/kronos-vs-qwen3-4b-instruct-2507.md): C 55.2 vs C 55.7\n- [Kronos vs Qwen3.5-0.8B](https://www.anchorterminal.com/compare/kronos-vs-qwen3-5-0-8b.md): C 55.2 vs D 46.7\n- [Kronos vs Qwen3.5-122B-A10B](https://www.anchorterminal.com/compare/kronos-vs-qwen3-5-122b-a10b.md): C 55.2 vs B 63.6\n- [Kronos vs Qwen3.5-2B](https://www.anchorterminal.com/compare/kronos-vs-qwen3-5-2b.md): C 55.2 vs D 50.5\n- [Kronos vs Qwen3.5-397B-A17B](https://www.anchorterminal.com/compare/kronos-vs-qwen3-5-397b-a17b.md): C 55.2 vs B 64.6\n- [Kronos vs Qwen3.5-4B](https://www.anchorterminal.com/compare/kronos-vs-qwen3-5-4b.md): C 55.2 vs C 61.2\n- [Kronos vs Qwen3.5-9B](https://www.anchorterminal.com/compare/kronos-vs-qwen3-5-9b.md): C 55.2 vs C 60.5\n- [Kronos vs Qwen3.6-27B](https://www.anchorterminal.com/compare/kronos-vs-qwen3-6-27b.md): C 55.2 vs B 67.2\n- [Kronos vs Qwen3.6-35B-A3B](https://www.anchorterminal.com/compare/kronos-vs-qwen3-6-35b-a3b.md): C 55.2 vs B 63.2\n- [Kronos vs Qwen3.8-27B](https://www.anchorterminal.com/compare/kronos-vs-qwen3-8-27b.md): C 55.2 vs B 68.6\n- [Kronos vs Qwen3.8-Flash-Next](https://www.anchorterminal.com/compare/kronos-vs-qwen3-8-flash-next.md): C 55.2 vs C 61.8\n- [Kronos vs Qwen3-Coder-30B-A3B-Instruct](https://www.anchorterminal.com/compare/kronos-vs-qwen3-coder-30b-a3b-instruct.md): C 55.2 vs C 55.7\n- [Kronos vs Qwen3-Coder-Next](https://www.anchorterminal.com/compare/kronos-vs-qwen3-coder-next.md): C 55.2 vs C 57.6\n- [Kronos vs Qwen3-Omni-30B-A3B-Instruct](https://www.anchorterminal.com/compare/kronos-vs-qwen3-omni-30b-a3b-instruct.md): C 55.2 vs E 41.8\n- [Kronos vs Qwen3-VL-2B-Instruct](https://www.anchorterminal.com/compare/kronos-vs-qwen3-vl-2b-instruct.md): C 55.2 vs C 61\n- [Kronos vs Qwen3-VL-30B-A3B-Instruct](https://www.anchorterminal.com/compare/kronos-vs-qwen3-vl-30b-a3b-instruct.md): C 55.2 vs C 58\n- [Kronos vs Qwen3-VL-32B-Instruct](https://www.anchorterminal.com/compare/kronos-vs-qwen3-vl-32b-instruct.md): C 55.2 vs C 56.4\n- [Kronos vs Qwen3-VL-4B-Instruct](https://www.anchorterminal.com/compare/kronos-vs-qwen3-vl-4b-instruct.md): C 55.2 vs D 52.6\n- [Kronos vs Qwen3-VL-8B-Instruct](https://www.anchorterminal.com/compare/kronos-vs-qwen3-vl-8b-instruct.md): C 55.2 vs C 54.9\n- [Kronos vs Qwopus3.8-27B-Flash-V2](https://www.anchorterminal.com/compare/kronos-vs-qwopus3-8-27b-flash-v2.md): C 55.2 vs C 61\n- [Kronos vs Recursive Flint](https://www.anchorterminal.com/compare/kronos-vs-recursive-flint.md): C 55.2 vs C 61\n- [Kronos vs SmolLM3-3B](https://www.anchorterminal.com/compare/kronos-vs-smollm3-3b.md): C 55.2 vs C 61\n- [Kronos vs Spark-X2.6-4B](https://www.anchorterminal.com/compare/kronos-vs-spark-x2-6-4b.md): C 55.2 vs C 61\n- [Kronos vs Swift 1.5 Qwen3.8-27B](https://www.anchorterminal.com/compare/kronos-vs-swift-1-5-qwen3-8-27b.md): C 55.2 vs C 61\n- [Kronos vs Swift 1.5 Qwen3.8-Flash-Next](https://www.anchorterminal.com/compare/kronos-vs-swift-1-5-qwen3-8-flash-next.md): C 55.2 vs D 52.3\n- [Kronos vs ThinkingCap (Qwen3.8 27B)](https://www.anchorterminal.com/compare/kronos-vs-thinkingcap-qwen3-8-27b.md): C 55.2 vs C 61\n\n## Verify this listing\n\nFor the vendor. The badge or a plain link to this page verifies the listing, from a page under github.com/shiyu-coder, or the README of github.com/shiyu-coder/Kronos. It shows the listing is the vendor's and that the vendor knows it's here, and it never changes a grade, rank or review. The vendor sends the page's address to `POST https://www.anchorterminal.com/api/v1/verify` as `{\"slug\": \"kronos\", \"url\": \"…\"}`, or calls the `verify_listing` tool at https://www.anchorterminal.com/mcp. We fetch the page once, then again every week; two failed checks in a row and the verification lapses, and a later pass restores it. What we check: https://www.anchorterminal.com/builders/index.md#verify\n\nHTML badge:\n\n```html\n\u003ca href=\"https://www.anchorterminal.com/tools/kronos\"\u003e\u003cimg src=\"https://www.anchorterminal.com/badges/kronos.svg\" alt=\"Kronos on Anchor Terminal\" height=\"20\"\u003e\u003c/a\u003e\n```\n\nMarkdown badge, for a README:\n\n```markdown\n[![Kronos on Anchor Terminal](https://www.anchorterminal.com/badges/kronos.svg)](https://www.anchorterminal.com/tools/kronos)\n```\n\nPlain link:\n\n```html\n\u003ca href=\"https://www.anchorterminal.com/tools/kronos\"\u003eKronos on Anchor Terminal\u003c/a\u003e\n```\n\n## Share this listing\n\nFor the vendor. Sharing assets for social media, two PNGs of 1200 × 630 that say Kronos is listed on Anchor Terminal, with the vendor's logo and this page's address and no grade or score.\n\n- Dark: https://www.anchorterminal.com/assets/share/kronos-dark.png\n- Light: https://www.anchorterminal.com/assets/share/kronos-light.png\n- Create a video and download its frames: https://www.anchorterminal.com/assets/listing-video/index.html?listing=kronos\n",
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "attributionRequired": true,
    "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-10-25",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.6",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run",
    "terms": "https://www.anchorterminal.com/terms/#data"
  },
  "page": {
    "breadcrumbs": [
      {
        "name": "Home",
        "url": "https://www.anchorterminal.com/"
      },
      {
        "name": "Terminal",
        "url": "https://www.anchorterminal.com/tools/"
      },
      {
        "name": "Open-weight models",
        "url": "https://www.anchorterminal.com/categories/open-weights"
      },
      {
        "name": "Kronos",
        "url": ""
      }
    ],
    "description": "Kronos is an open-weight family of small forecasting models for financial candlestick (K-line) series, from Tsinghua University researchers. It reads open, high, low, close, volume and amount bars with timestamps and generates forecast bars, under MIT.",
    "facts": [
      "#68 in Open weights",
      "None auth",
      "0 desk reviews"
    ],
    "h1": "Kronos",
    "image": "https://www.anchorterminal.com/assets/og/tools-kronos.png",
    "path": "/tools/kronos",
    "published": "2026-10-01",
    "section": "tools",
    "title": "Kronos review (2026): pricing, alternatives and grade C",
    "toc": null,
    "updated": "2026-10-25",
    "url": "https://www.anchorterminal.com/tools/kronos"
  },
  "tokens": {
    "markdown": 8650,
    "slim": 2280
  },
  "version": 1
}