{
  "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": "nanbeige-4-3-33b",
      "name": "Nanbeige4.3-33B",
      "vendor": "Nanbeige LLM Lab (BOSS Zhipin)",
      "vendorUrl": "https://huggingface.co/Nanbeige",
      "kind": "weights",
      "category": "open-weights",
      "summary": "Nanbeige4.3-33B is the Nanbeige LLM Lab's open-weight dense model for agentic coding and office work, 33B parameters on the Llama architecture with a 512K-token context, thinking that switches per request and tool calls in an XML format.",
      "url": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b",
      "markdownUrl": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b.md",
      "slimMarkdownUrl": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b.min.md",
      "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/nanbeige-4-3-33b.json",
      "repo": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B",
      "license": "Apache-2.0 (Hub metadata only; the repository has no LICENSE file)",
      "transports": [],
      "packages": [
        {
          "registry": "huggingface",
          "name": "Nanbeige/Nanbeige4.3-33B"
        }
      ],
      "auth": "none",
      "authNotes": "The weights download from Hugging Face with no account, token or terms to accept. The Hub reports gated false for Nanbeige/Nanbeige4.3-33B and Nanbeige/Nanbeige4.3-33B-Base (checked 2026-10-25).",
      "pricing": "free",
      "pricingNotes": "Free to download and use, commercial use included, under Apache 2.0 as the card's metadata names it. Neither repository holds a LICENSE file. Running it costs the hardware, 66.26 GB of BF16 files and no quantised build from the publisher. No hosted API or inference provider serves it (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": null,
        "npmWeekly": null,
        "pypiWeekly": null,
        "asOf": "2026-10-25"
      },
      "docsUrl": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B",
      "capabilities": [
        "model.open-weights"
      ],
      "tags": [
        "open-weights",
        "model",
        "dense",
        "llama",
        "reasoning",
        "tool-calling",
        "coding",
        "agentic",
        "long-context",
        "safetensors",
        "self-hosted",
        "new"
      ],
      "lastRelease": "2026-10-24",
      "graded": true,
      "anchor": {
        "graded": true,
        "score": 61,
        "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": 1211,
        "categoryRank": 48,
        "methodology": "0.6",
        "run": "2026-10-01",
        "scores": {
          "interface": 85,
          "reach": 55,
          "risk": 60,
          "upkeep": 56
        },
        "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 earns 10, since the only build from Nanbeige or a runtime's library is 66.26 GB of BF16, about 80 GB with a fifth added, which fits the reference machine but not 64 GB. Runtimes earn 20 from released vLLM, SGLang, Transformers and llama.cpp, which all run the Llama architecture, and Ollama and LM Studio have no build. Builds earn 0, as Nanbeige publishes no quantised build. Terms earn 15 for Apache 2.0 with nothing added, though only the metadata names it. Download earns 10, with no gate. Hosted earns 0, as no provider or Nanbeige API serves it.",
            "checked": "2026-10-25"
          },
          {
            "key": "risk",
            "name": "Risk",
            "weight": 20,
            "effectiveWeight": 30.77,
            "score": 60,
            "points": 18.46,
            "reason": "File safety earns 25, safetensors only and no repository code, so the card's trust_remote_code has nothing to run. Provenance earns 15, the lab's own organisation as its arXiv report links it, with a hash for every file. Safety evaluation earns 0 and agent guidance 0, since neither card says anything on either. Licence risk keeps all 15 under Apache 2.0. The security programme earns 5 for BOSS Zhipin's security response centre, which isn't linked to the model.",
            "checked": "2026-10-25"
          },
          {
            "key": "interface",
            "name": "Agent interface",
            "weight": 10,
            "effectiveWeight": 15.38,
            "score": 85,
            "points": 13.08,
            "reason": "The chat template earns 20, Jinja in tokenizer_config.json. The tool-calling format earns 25, with the XML call format in the template and the qwen3_coder parser named for vLLM and SGLang. Trained tool calling earns 20, several calls per turn in the template. Structured output earns 0, as nothing mentions it. Reasoning control earns 10, thinking switched per request with the qwen3 reasoning parser named. Context earns 10 for 512K set in the config with no scaling.",
            "checked": "2026-10-25"
          },
          {
            "key": "upkeep",
            "name": "Documentation and upkeep",
            "weight": 10,
            "effectiveWeight": 15.38,
            "score": 56,
            "points": 8.62,
            "reason": "The model card earns 11, with intended use, parameters, context and languages but no limitations, architecture or cut-off. Training data earns 0, as nothing describes it. Last change earns 20 for 25 October 2026. Responsiveness earns 10, a Nanbeige member answering four of ten discussions on release day. Version policy earns 5 for a first release, with a config change made and reverted under bare commit titles. File health earns 10, the files and the card agreeing.",
            "checked": "2026-10-25"
          }
        ],
        "assessment": {
          "date": "2026-10-25",
          "basis": "public evidence",
          "confidence": "low",
          "notes": {
            "capability": "Pending for this listing. Independent results were found in 0 areas (none), and Capability needs 2. Its weight is shared across the other four categories and the total is capped at 61 until there are more, or our own runs start.",
            "interface": "The chat template earns 20, Jinja in tokenizer_config.json. The tool-calling format earns 25, with the XML call format in the template and the qwen3_coder parser named for vLLM and SGLang. Trained tool calling earns 20, several calls per turn in the template. Structured output earns 0, as nothing mentions it. Reasoning control earns 10, thinking switched per request with the qwen3 reasoning parser named. Context earns 10 for 512K set in the config with no scaling.",
            "reach": "Memory earns 10, since the only build from Nanbeige or a runtime's library is 66.26 GB of BF16, about 80 GB with a fifth added, which fits the reference machine but not 64 GB. Runtimes earn 20 from released vLLM, SGLang, Transformers and llama.cpp, which all run the Llama architecture, and Ollama and LM Studio have no build. Builds earn 0, as Nanbeige publishes no quantised build. Terms earn 15 for Apache 2.0 with nothing added, though only the metadata names it. Download earns 10, with no gate. Hosted earns 0, as no provider or Nanbeige API serves it.",
            "risk": "File safety earns 25, safetensors only and no repository code, so the card's trust_remote_code has nothing to run. Provenance earns 15, the lab's own organisation as its arXiv report links it, with a hash for every file. Safety evaluation earns 0 and agent guidance 0, since neither card says anything on either. Licence risk keeps all 15 under Apache 2.0. The security programme earns 5 for BOSS Zhipin's security response centre, which isn't linked to the model.",
            "upkeep": "The model card earns 11, with intended use, parameters, context and languages but no limitations, architecture or cut-off. Training data earns 0, as nothing describes it. Last change earns 20 for 25 October 2026. Responsiveness earns 10, a Nanbeige member answering four of ten discussions on release day. Version policy earns 5 for a first release, with a config change made and reverted under bare commit titles. File health earns 10, the files and the card agreeing."
          },
          "sources": [
            {
              "what": "model card, benchmarks, quickstart, parsers, sampling settings (read 25 October 2026)",
              "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B",
              "seen": "2026-10-25"
            },
            {
              "what": "Hub API, gated false, licence tag, base model, parameter count, empty provider mapping",
              "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B",
              "seen": "2026-10-25"
            },
            {
              "what": "file tree, sizes and SHA-256, no LICENSE or Python files",
              "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/tree/main?recursive=true",
              "seen": "2026-10-25"
            },
            {
              "what": "commit history, upload on 24 October, config change and revert, card on 25 October",
              "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/commits/main",
              "seen": "2026-10-25"
            },
            {
              "what": "ten discussions, authors and dates only",
              "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/discussions",
              "seen": "2026-10-25"
            },
            {
              "what": "LlamaForCausalLM, 524,288 positions, rope_theta, use_cache",
              "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B/raw/main/config.json",
              "seen": "2026-10-25"
            },
            {
              "what": "chat template, tool and thinking switches, default system prompts",
              "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B/raw/main/tokenizer_config.json",
              "seen": "2026-10-25"
            },
            {
              "what": "generation config, use_cache false",
              "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B/raw/main/generation_config.json",
              "seen": "2026-10-25"
            },
            {
              "what": "Base model card and files, 131,072 context, Apache 2.0 tag",
              "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base",
              "seen": "2026-10-25"
            },
            {
              "what": "organisation Nanbeige LLM Lab, not verified",
              "url": "https://huggingface.co/api/organizations/Nanbeige/overview",
              "seen": "2026-10-25"
            },
            {
              "what": "Nanbeige4.2-3B report, Nanbeige LLM Lab, Boss Zhipin, link to the Hub organisation",
              "url": "https://arxiv.org/html/2607.22083",
              "seen": "2026-10-25"
            },
            {
              "what": "vLLM v0.32.0 LlamaForCausalLM (parsers in vllm/tool_parsers and vllm/reasoning)",
              "url": "https://raw.githubusercontent.com/vllm-project/vllm/v0.32.0/vllm/model_executor/models/registry.py",
              "seen": "2026-10-25"
            },
            {
              "what": "SGLang v0.5.22 qwen3_coder parser (llama.py and the qwen3 reasoning parser at the same tag)",
              "url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.22/python/sglang/srt/function_call/function_call_parser.py",
              "seen": "2026-10-25"
            },
            {
              "what": "Transformers v5.20.2 llama",
              "url": "https://raw.githubusercontent.com/huggingface/transformers/v5.20.2/src/transformers/models/llama/modeling_llama.py",
              "seen": "2026-10-25"
            },
            {
              "what": "llama.cpp b11888 converter, LlamaForCausalLM mapped to llama",
              "url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/b11888/conversion/__init__.py",
              "seen": "2026-10-25"
            },
            {
              "what": "Ollama search, no library entry for Nanbeige4.3",
              "url": "https://ollama.com/search?q=nanbeige",
              "seen": "2026-10-25"
            },
            {
              "what": "OpenRouter model list, no Nanbeige model",
              "url": "https://openrouter.ai/api/v1/models",
              "seen": "2026-10-25"
            },
            {
              "what": "BOSS Zhipin security response centre",
              "url": "https://security.zhipin.com/",
              "seen": "2026-10-25"
            },
            {
              "what": "Artificial Analysis leaderboard, Nanbeige4.1-3B only",
              "url": "https://artificialanalysis.ai/models",
              "seen": "2026-10-25"
            }
          ],
          "openQuestions": [
            "Capability is computed later. No independent result was found on 25 October 2026 (Artificial Analysis, Vals AI, Scale SEAL, Terminal-Bench, SWE-bench, BFCL); LMArena answered 403. So Capability stays pending and the total is capped at 61",
            "The repositories have no LICENSE file. Apache 2.0 rests on the metadata tag, scored by its terms under the calibration. Worth asking Nanbeige to add the file",
            "Who makes it is established from the lab's arXiv report (Boss Zhipin), the template's default system prompt and the card's kanzhun.com contact. That Kanzhun Limited is the legal entity wasn't confirmed first-hand, as ir.zhipin.com refused our reader",
            "Context earns 10 on the config's 524,288 tokens with no scaling. Nanbeige doesn't say the trained length; if 512K was reached by raising rope_theta without long training, a stricter reading gives 5 and Interface 80",
            "Tool-calling format earns 25 on the template's XML format and the named qwen3_coder parser. The card shows no returned call; reading that as half the requirement gives 10 and Interface 70",
            "Reasoning control earns 10 on enable_thinking and the named qwen3 parser. The card doesn't say in words where the reasoning comes back; a stricter reading gives 0",
            "Version policy 5. The config change and revert on 24 and 25 October carry bare commit titles; reading them as files replaced without a note gives 0 and Upkeep 51",
            "Responsiveness counted on ten discussions all opened on release day; recheck at the next pass",
            "The weights date is 24 October 2026 by the upload commit. The Hub doesn't show when the repositories turned public; the first discussion is from 12:36 UTC on 25 October",
            "unchecked, the registration date of zhipin.com"
          ]
        },
        "negative": 0,
        "verdict": "A dense 33B Llama-architecture cousin for long agentic coding and office tasks, with per-request thinking, XML tool calls that stock vLLM and SGLang parse, and 512K of context. It needs about 96 GB at BF16, as Nanbeige ships no quantised build. Its agentic results so far are Nanbeige's own, with no safety results, training data notes or licence file.",
        "bestFor": "Agents with about 96 GB of GPU memory, or the 128 GB reference machine, that want an Apache-tagged 33B model for long coding and tool loops and can wait for independent results.",
        "strengths": [
          "A plain LlamaForCausalLM with no repository code, so released vLLM, SGLang, Transformers and llama.cpp load it",
          "Apache 2.0 by its metadata, ungated, safetensors only",
          "524,288 tokens of context set in config.json with no rope scaling",
          "Thinking switches per request, and preserve_thinking keeps earlier reasoning for agent loops",
          "Tool calls in an XML format the qwen3_coder parser reads, several per turn",
          "A Nanbeige member answered four of the first ten discussions on release day"
        ],
        "weaknesses": [
          "No independent benchmark result yet; every agentic number on the card is Nanbeige's own run",
          "BF16 only from Nanbeige, about 80 GB to load, and no Ollama or LM Studio build",
          "No LICENSE file in either repository; Apache 2.0 rests on the metadata tag",
          "No safety evaluation, agent-use guidance, training data description or knowledge cut-off",
          "No hosted API or inference provider serves it"
        ],
        "agentNotes": [
          "Serve with `vllm serve Nanbeige/Nanbeige4.3-33B --enable-auto-tool-choice --reasoning-parser qwen3 --tool-call-parser qwen3_coder`, as the card says, or SGLang with the same two parsers",
          "Keep `tool_call_format` at its default `xml`, since qwen3_coder parses that format; the `json` option writes Hermes-style JSON calls, which need a different parser such as vLLM's hermes",
          "Pass `chat_template_kwargs` `{\"preserve_thinking\": true}` for multi-turn tool use and coding, and `{\"enable_thinking\": false}` to skip reasoning, as the card recommends",
          "Use temperature 1.0, top_p 0.95 and up to 65,536 new tokens for agent work, and temperature 0.6 for reasoning and chat, per the card",
          "Send your own system prompt, since without one the template injects a Chinese identity line, or a Chinese tool-expert prompt when tools are passed",
          "Send one system message at the start only, as the template raises an error for a later one",
          "In plain Transformers pass `use_cache=True` to generate, since config.json and generation_config.json set it false; trust_remote_code isn't needed"
        ],
        "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": 61
          }
        ],
        "editorialScores": {
          "interface": 85,
          "reach": 55,
          "risk": 60,
          "upkeep": 56
        },
        "provenanceScore": 53,
        "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": 61
          }
        ]
      },
      "connect": {
        "install": "# vLLM 0.32.0 or SGLang 0.5.22, as the card's commands (the card uses four GPUs with tensor parallelism)\nhf download Nanbeige/Nanbeige4.3-33B --local-dir Nanbeige4.3-33B\nvllm serve ./Nanbeige4.3-33B --served-model-name nanbeige --tensor-parallel-size 4 \\\n  --enable-auto-tool-choice --reasoning-parser qwen3 --tool-call-parser qwen3_coder\n# or\npython -m sglang.launch_server --model-path ./Nanbeige4.3-33B --tp-size 4 \\\n  --reasoning-parser qwen3 --tool-call-parser qwen3_coder",
        "http": "curl http://localhost:8000/v1/chat/completions -H 'Content-Type: application/json' \\\n  -d '{\"model\": \"nanbeige\", \"messages\": [{\"role\": \"user\", \"content\": \"Which number is bigger, 9.11 or 9.8?\"}],\n       \"temperature\": 0.6, \"top_p\": 0.95, \"chat_template_kwargs\": {\"enable_thinking\": true}}'"
      },
      "letme": {
        "capability": "https://letme.dev/model.open-weights",
        "tool": "https://letme.dev/nanbeige-4-3-33b"
      },
      "weights": {
        "trainedBy": "Nanbeige LLM Lab (BOSS Zhipin)",
        "builtOn": "Nanbeige/Nanbeige4.3-33B-Base (fine-tune, declared in the card's metadata; Apache 2.0 by its tag, no LICENSE file)",
        "params": "33B",
        "context": 524288,
        "goodFor": "A cousin for long agentic coding and office tasks on a machine with about 96 GB of memory, if Nanbeige's own results hold up, with a 512K context and per-request thinking.",
        "runtimes": [
          "vLLM",
          "SGLang",
          "Transformers",
          "llama.cpp"
        ],
        "builds": [
          {
            "name": "Nanbeige4.3-33B (BF16)",
            "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B",
            "who": "publisher",
            "precision": "bf16",
            "sizeGB": 66.26,
            "memoryGB": 96,
            "format": "safetensors",
            "runsOn": [
              "vLLM",
              "SGLang",
              "Transformers"
            ],
            "note": "Whole repository, 66.24 GB of weights in 14 shards. 66.26 GB × 1.2 = 79.5 GB. The only build from Nanbeige; no Ollama, LM Studio or ggml-org build exists, and llama.cpp needs a GGUF converted with its own script or taken from someone else"
          }
        ],
        "run": "vllm serve Nanbeige/Nanbeige4.3-33B --tensor-parallel-size 4 --enable-auto-tool-choice --reasoning-parser qwen3 --tool-call-parser qwen3_coder",
        "runBuild": "Nanbeige4.3-33B (BF16)",
        "checked": "2026-10-25",
        "minMemoryGB": 96
      },
      "family": {
        "lab": "Nanbeige LLM Lab (BOSS Zhipin)",
        "family": "Nanbeige4.3",
        "url": "https://www.anchorterminal.com/open-weights/#nanbeige",
        "others": [
          {
            "name": "Nanbeige4.3-33B-Base",
            "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base",
            "params": "33B"
          }
        ],
        "olderCount": 7
      },
      "notable": [
        "Weights uploaded on 24 October 2026. Both repositories were created at 02:46 and 02:49 UTC that day, the BF16 shards went in at 11:57 UTC, and the full card was written on 25 October. The Hub doesn't record when the repositories turned public; the first discussion and the first copy by someone else both date from about 12:40 UTC on 25 October. On 25 October the Hub counted 344 likes (https://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/commits/main)",
        "Who makes it. The organisation is Nanbeige LLM Lab on Hugging Face (not verified by the Hub). The lab's Nanbeige4.2-3B report on arXiv gives its affiliation as Boss Zhipin and links the same organisation, this model's chat template tells the model it was developed and trained by BOSS Zhipin when no system prompt is given, and the card's contact address is on kanzhun.com, a site that loads BOSS Zhipin's assets (https://arxiv.org/abs/2607.22083)",
        "Nanbeige calls it the flagship teacher of the Nanbeige4.3 series, to be distilled into smaller models starting with Nanbeige4.3-3B, which isn't released (https://huggingface.co/Nanbeige/Nanbeige4.3-33B)",
        "Nanbeige's card claims the best agentic results among open models around 30B, ahead of Qwen3.8-27B, Muse-Glimmer-30B and Gemma4-31B on 15 of 17 agentic rows in its own runs, among them Terminal-Bench 2.1 80.9 with Terminus 2, DeepSWE 65.5, Toolathlon-Verified 70.4 and Tau3-Bench 48.8, with Qwen3.8-27B ahead on SWE-Bench Pro and GDPval. Some rows were run in Claude Code and mini-swe-agent harnesses. These are the publisher's numbers, and no independent result was found on 25 October 2026",
        "Fine-tuned from Nanbeige4.3-33B-Base, published the same day in its own repository with the same 52-layer Llama shape, a 131,072-token context, a plain-text template and no discussions yet. Its card compares it only with Gemma4-31B-Base, in Nanbeige's own runs (https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base)",
        "config.json was switched to a 1,048,576-token context with YaRN at 11:58 UTC on 24 October and back to 524,288 tokens with no scaling at 02:44 UTC on 25 October, both under the commit title Update config.json (https://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/commits/main)",
        "Nanbeige4.2-3B, the previous release, is a looped transformer with modeling code in its repository and Nanbeige forks of vLLM, SGLang, llama.cpp and Ollama on its card. 4.3-33B is a plain LlamaForCausalLM with no repository code, so stock runtimes load it. Within a day others had posted more than 40 GGUF, MLX, FP8, NVFP4 and EXL3 copies, none from Nanbeige or a runtime's own library"
      ],
      "area": "models",
      "details": [
        {
          "label": "Surface graded",
          "value": "The open weights at huggingface.co/Nanbeige/Nanbeige4.3-33B, read from the card, the files and their metadata, the chat template and the runtimes' own code on 25 October 2026, with the Base repository as a note. Nothing was downloaded or run"
        },
        {
          "label": "Model",
          "value": "LlamaForCausalLM, 33,118,977,024 BF16 parameters by the Hub's count, 52 layers, hidden size 7,168, 56 attention heads with 8 KV heads (grouped-query attention), head size 128, vocabulary 166,912, untied embeddings. Standard Llama tensor names, no custom code and no auto_map. Text in and text out; the template answers images, audio and video with a note that the model can't read them"
        },
        {
          "label": "Files",
          "value": "66.26 GB in all, 14 BF16 safetensors shards (66.24 GB), tokenizer.json and a SentencePiece tokenizer.model, tokenizer_config.json with the chat template, configs and three figures. No pickle files, no Python files, no LICENSE file. SHA-256 on every LFS file"
        },
        {
          "label": "Licence",
          "value": "Apache 2.0 in the card's metadata (license apache-2.0) and on the Base card. Neither repository has a LICENSE file and the card text has no licence section, so the terms rest on the metadata tag. No use policy or other terms found"
        },
        {
          "label": "Runtimes",
          "value": "The card gives Transformers, vLLM and SGLang commands. Released vLLM 0.32.0, SGLang 0.5.22, Transformers 5.20.2 and llama.cpp b11888 all carry LlamaForCausalLM, and vLLM and SGLang carry the qwen3 reasoning parser and qwen3_coder tool parser the card names. No Ollama library entry, no LM Studio catalogue entry and no ggml-org or lmstudio-community build on 25 October 2026"
        },
        {
          "label": "Context",
          "value": "524,288 tokens, max_position_embeddings in config.json with rope_theta 120,000,000 and no rope scaling, and the card's 512K. The Base model is 131,072 tokens with rope_theta 10,000,000. The card doesn't say at what length the model was trained"
        },
        {
          "label": "Tools and reasoning",
          "value": "Passing tools switches the template to a tool prompt. tool_call_format xml (the default, which Nanbeige recommends) asks for \u003ctool_call\u003e\u003cfunction=name\u003e\u003cparameter=...\u003e blocks, read back by the qwen3_coder parser in vLLM and SGLang; json asks for Hermes-style JSON in \u003ctool_call\u003e tags. The template renders several calls per turn and groups tool results into one user turn. enable_thinking switches thinking per request (on by default), the qwen3 reasoning parser splits it out, and preserve_thinking keeps earlier turns' reasoning, which Nanbeige recommends for agent and coding work. No thinking budget documented"
        },
        {
          "label": "Hardware",
          "value": "Not stated. The card's vLLM and SGLang commands use tensor parallelism of 4. The BF16 files need about 80 GB with a fifth added, so 96 GB of memory, within the 128 GB reference machine's GPU share"
        },
        {
          "label": "Training data",
          "value": "Not described on either card, with no cut-off. Nanbeige's earlier technical reports covered earlier models only"
        }
      ],
      "provenance": {
        "legalEntity": "BOSS Zhipin (Kanzhun Limited), unconfirmed",
        "domain": "zhipin.com",
        "domainRegistered": "",
        "endpointOnVendorDomain": null,
        "terms": "",
        "privacy": "",
        "statusPage": "",
        "changelog": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B/commits/main",
        "securityTxt": "none",
        "checked": "2026-10-25",
        "notes": [
          "The card names the Nanbeige Team and nothing else. BOSS Zhipin is identified as the maker by the lab's Nanbeige4.2-3B report on arXiv (Nanbeige LLM Lab, Boss Zhipin), which links huggingface.co/Nanbeige, and by the chat template's default system prompt naming BOSS Zhipin as developer and trainer. The card's contact address is on kanzhun.com. Kanzhun Limited as BOSS Zhipin's operator wasn't confirmed first-hand, since ir.zhipin.com refused our reader. The domain's registration date wasn't checked.",
          "The Hugging Face organisation Nanbeige (Nanbeige LLM Lab) isn't verified by the Hub. The arXiv report's link to it is the publisher's own documentation pointing there.",
          "Nothing is hosted, so the licence stands in for terms of service and no privacy policy is scored.",
          "The changelog is the repository's commit history.",
          "www.zhipin.com/.well-known/security.txt answers 404 and kanzhun.com's redirects to the home page. BOSS Zhipin runs a security response centre at security.zhipin.com for reporting flaws in its products; nothing links it to the model."
        ],
        "score": 53,
        "checks": [
          {
            "check": "Legal entity named",
            "value": "BOSS Zhipin (Kanzhun Limited), unconfirmed",
            "points": 20,
            "max": 20,
            "state": "ok"
          },
          {
            "check": "Domain age",
            "value": "zhipin.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 Apache-2.0 (Hub metadata only; the repository has no LICENSE file) 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": "published",
            "points": 10,
            "max": 10,
            "state": "ok"
          },
          {
            "check": "security.txt",
            "value": "not found",
            "points": 0,
            "max": 10,
            "state": "no"
          }
        ]
      },
      "pageJsonUrl": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b.json"
    },
    "verify": {
      "accepts": "a page on zhipin.com or one of its subdomains, or a page under huggingface.co/nanbeige",
      "badgeUrl": "https://www.anchorterminal.com/badges/nanbeige-4-3-33b.svg",
      "body": {
        "slug": "nanbeige-4-3-33b",
        "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/nanbeige-4-3-33b",
      "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/nanbeige-4-3-33b\"\u003e\u003cimg src=\"https://www.anchorterminal.com/badges/nanbeige-4-3-33b.svg\" alt=\"Nanbeige4.3-33B on Anchor Terminal\" height=\"20\"\u003e\u003c/a\u003e",
        "markdown": "[![Nanbeige4.3-33B on Anchor Terminal](https://www.anchorterminal.com/badges/nanbeige-4-3-33b.svg)](https://www.anchorterminal.com/tools/nanbeige-4-3-33b)",
        "link": "\u003ca href=\"https://www.anchorterminal.com/tools/nanbeige-4-3-33b\"\u003eNanbeige4.3-33B 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/nanbeige-4-3-33b",
    "json": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b.json",
    "llms": "https://www.anchorterminal.com/llms.txt",
    "markdown": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b.md",
    "mcp": "https://www.anchorterminal.com/mcp",
    "pricing": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b#pricing",
    "reviews": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b#reviews",
    "slim": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b.min.md"
  },
  "markdown": "## Overview\n\n**Grade C · 61/100 · rank #56 of 111 open-weight models · #48 in Open-weight models · not agent-ready · confidence low**\n\n\nTrained by Nanbeige LLM Lab (BOSS Zhipin), in the Nanbeige4.3 family, with [Nanbeige4.3-33B-Base](https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base). Every model from this lab and its older versions: https://www.anchorterminal.com/open-weights/#nanbeige.\n\n## Assessment\n\nA dense 33B Llama-architecture cousin for long agentic coding and office tasks, with per-request thinking, XML tool calls that stock vLLM and SGLang parse, and 512K of context. It needs about 96 GB at BF16, as Nanbeige ships no quantised build. Its agentic results so far are Nanbeige's own, with no safety results, training data notes or licence file.\n\n## Facts\n\n| Field | Value |\n| --- | --- |\n| Vendor | Nanbeige LLM Lab (BOSS Zhipin) (https://huggingface.co/Nanbeige) |\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 Nanbeige/Nanbeige4.3-33B and Nanbeige/Nanbeige4.3-33B-Base (checked 2026-10-25). |\n| Pricing | Free (Free) · Free to download and use, commercial use included, under Apache 2.0 as the card's metadata names it. Neither repository holds a LICENSE file. Running it costs the hardware, 66.26 GB of BF16 files and no quantised build from the publisher. No hosted API or inference provider serves it (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 | Apache-2.0 (Hub metadata only; the repository has no LICENSE file) |\n| Packages | huggingface: `Nanbeige/Nanbeige4.3-33B` |\n| Source | https://huggingface.co/Nanbeige/Nanbeige4.3-33B |\n| Docs | https://huggingface.co/Nanbeige/Nanbeige4.3-33B |\n| llms.txt | not found |\n| Last release | 2026-10-24 |\n| Surface graded | The open weights at huggingface.co/Nanbeige/Nanbeige4.3-33B, read from the card, the files and their metadata, the chat template and the runtimes' own code on 25 October 2026, with the Base repository as a note. Nothing was downloaded or run |\n| Model | LlamaForCausalLM, 33,118,977,024 BF16 parameters by the Hub's count, 52 layers, hidden size 7,168, 56 attention heads with 8 KV heads (grouped-query attention), head size 128, vocabulary 166,912, untied embeddings. Standard Llama tensor names, no custom code and no auto_map. Text in and text out; the template answers images, audio and video with a note that the model can't read them |\n| Files | 66.26 GB in all, 14 BF16 safetensors shards (66.24 GB), tokenizer.json and a SentencePiece tokenizer.model, tokenizer_config.json with the chat template, configs and three figures. No pickle files, no Python files, no LICENSE file. SHA-256 on every LFS file |\n| Licence | Apache 2.0 in the card's metadata (license apache-2.0) and on the Base card. Neither repository has a LICENSE file and the card text has no licence section, so the terms rest on the metadata tag. No use policy or other terms found |\n| Runtimes | The card gives Transformers, vLLM and SGLang commands. Released vLLM 0.32.0, SGLang 0.5.22, Transformers 5.20.2 and llama.cpp b11888 all carry LlamaForCausalLM, and vLLM and SGLang carry the qwen3 reasoning parser and qwen3_coder tool parser the card names. No Ollama library entry, no LM Studio catalogue entry and no ggml-org or lmstudio-community build on 25 October 2026 |\n| Context | 524,288 tokens, max_position_embeddings in config.json with rope_theta 120,000,000 and no rope scaling, and the card's 512K. The Base model is 131,072 tokens with rope_theta 10,000,000. The card doesn't say at what length the model was trained |\n| Tools and reasoning | Passing tools switches the template to a tool prompt. tool_call_format xml (the default, which Nanbeige recommends) asks for \u003ctool_call\u003e\u003cfunction=name\u003e\u003cparameter=...\u003e blocks, read back by the qwen3_coder parser in vLLM and SGLang; json asks for Hermes-style JSON in \u003ctool_call\u003e tags. The template renders several calls per turn and groups tool results into one user turn. enable_thinking switches thinking per request (on by default), the qwen3 reasoning parser splits it out, and preserve_thinking keeps earlier turns' reasoning, which Nanbeige recommends for agent and coding work. No thinking budget documented |\n| Hardware | Not stated. The card's vLLM and SGLang commands use tensor parallelism of 4. The BF16 files need about 80 GB with a fifth added, so 96 GB of memory, within the 128 GB reference machine's GPU share |\n| Training data | Not described on either card, with no cut-off. Nanbeige's earlier technical reports covered earlier models only |\n| Capabilities | model.open-weights |\n| Tags | open-weights, model, dense, llama, reasoning, tool-calling, coding, agentic, long-context, safetensors, self-hosted, new |\n| JSON | https://www.anchorterminal.com/api/v1/tools/nanbeige-4-3-33b.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 | 60 | 18.5 |\n| Agent interface | 10% | 15.4 | 85 | 13.1 |\n| Documentation and upkeep | 10% | 15.4 | 56 | 8.6 |\n| Negative events | up to −15 | up to −15 | none recorded | 0 |\n| **Total** | | | | **61 → 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 earns 10, since the only build from Nanbeige or a runtime's library is 66.26 GB of BF16, about 80 GB with a fifth added, which fits the reference machine but not 64 GB. Runtimes earn 20 from released vLLM, SGLang, Transformers and llama.cpp, which all run the Llama architecture, and Ollama and LM Studio have no build. Builds earn 0, as Nanbeige publishes no quantised build. Terms earn 15 for Apache 2.0 with nothing added, though only the metadata names it. Download earns 10, with no gate. Hosted earns 0, as no provider or Nanbeige API serves it.\n- Risk 60 (checked 2026-10-25): File safety earns 25, safetensors only and no repository code, so the card's trust_remote_code has nothing to run. Provenance earns 15, the lab's own organisation as its arXiv report links it, with a hash for every file. Safety evaluation earns 0 and agent guidance 0, since neither card says anything on either. Licence risk keeps all 15 under Apache 2.0. The security programme earns 5 for BOSS Zhipin's security response centre, which isn't linked to the model.\n- Agent interface 85 (checked 2026-10-25): The chat template earns 20, Jinja in tokenizer_config.json. The tool-calling format earns 25, with the XML call format in the template and the qwen3_coder parser named for vLLM and SGLang. Trained tool calling earns 20, several calls per turn in the template. Structured output earns 0, as nothing mentions it. Reasoning control earns 10, thinking switched per request with the qwen3 reasoning parser named. Context earns 10 for 512K set in the config with no scaling.\n- Documentation and upkeep 56 (checked 2026-10-25): The model card earns 11, with intended use, parameters, context and languages but no limitations, architecture or cut-off. Training data earns 0, as nothing describes it. Last change earns 20 for 25 October 2026. Responsiveness earns 10, a Nanbeige member answering four of ten discussions on release day. Version policy earns 5 for a first release, with a config change made and reverted under bare commit titles. File health earns 10, the files and the card agreeing.\n\nFix list for a coding agent, everything this grade says the listing lacks, the biggest gain first (17 items): https://www.anchorterminal.com/fixes/nanbeige-4-3-33b.md (JSON https://www.anchorterminal.com/fixes/nanbeige-4-3-33b.json)\n\n### What we couldn't check\n\n- Capability is computed later. No independent result was found on 25 October 2026 (Artificial Analysis, Vals AI, Scale SEAL, Terminal-Bench, SWE-bench, BFCL); LMArena answered 403. So Capability stays pending and the total is capped at 61\n- The repositories have no LICENSE file. Apache 2.0 rests on the metadata tag, scored by its terms under the calibration. Worth asking Nanbeige to add the file\n- Who makes it is established from the lab's arXiv report (Boss Zhipin), the template's default system prompt and the card's kanzhun.com contact. That Kanzhun Limited is the legal entity wasn't confirmed first-hand, as ir.zhipin.com refused our reader\n- Context earns 10 on the config's 524,288 tokens with no scaling. Nanbeige doesn't say the trained length; if 512K was reached by raising rope_theta without long training, a stricter reading gives 5 and Interface 80\n- Tool-calling format earns 25 on the template's XML format and the named qwen3_coder parser. The card shows no returned call; reading that as half the requirement gives 10 and Interface 70\n- Reasoning control earns 10 on enable_thinking and the named qwen3 parser. The card doesn't say in words where the reasoning comes back; a stricter reading gives 0\n- Version policy 5. The config change and revert on 24 and 25 October carry bare commit titles; reading them as files replaced without a note gives 0 and Upkeep 51\n- Responsiveness counted on ten discussions all opened on release day; recheck at the next pass\n- The weights date is 24 October 2026 by the upload commit. The Hub doesn't show when the repositories turned public; the first discussion is from 12:36 UTC on 25 October\n- unchecked, the registration date of zhipin.com\n\n### Sources\n\n- model card, benchmarks, quickstart, parsers, sampling settings (read 25 October 2026): \u003chttps://huggingface.co/Nanbeige/Nanbeige4.3-33B\u003e (seen 2026-10-25)\n- Hub API, gated false, licence tag, base model, parameter count, empty provider mapping: \u003chttps://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B\u003e (seen 2026-10-25)\n- file tree, sizes and SHA-256, no LICENSE or Python files: \u003chttps://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/tree/main?recursive=true\u003e (seen 2026-10-25)\n- commit history, upload on 24 October, config change and revert, card on 25 October: \u003chttps://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/commits/main\u003e (seen 2026-10-25)\n- ten discussions, authors and dates only: \u003chttps://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/discussions\u003e (seen 2026-10-25)\n- LlamaForCausalLM, 524,288 positions, rope_theta, use_cache: \u003chttps://huggingface.co/Nanbeige/Nanbeige4.3-33B/raw/main/config.json\u003e (seen 2026-10-25)\n- chat template, tool and thinking switches, default system prompts: \u003chttps://huggingface.co/Nanbeige/Nanbeige4.3-33B/raw/main/tokenizer_config.json\u003e (seen 2026-10-25)\n- generation config, use_cache false: \u003chttps://huggingface.co/Nanbeige/Nanbeige4.3-33B/raw/main/generation_config.json\u003e (seen 2026-10-25)\n- Base model card and files, 131,072 context, Apache 2.0 tag: \u003chttps://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base\u003e (seen 2026-10-25)\n- organisation Nanbeige LLM Lab, not verified: \u003chttps://huggingface.co/api/organizations/Nanbeige/overview\u003e (seen 2026-10-25)\n- Nanbeige4.2-3B report, Nanbeige LLM Lab, Boss Zhipin, link to the Hub organisation: \u003chttps://arxiv.org/html/2607.22083\u003e (seen 2026-10-25)\n- vLLM v0.32.0 LlamaForCausalLM (parsers in vllm/tool_parsers and vllm/reasoning): \u003chttps://raw.githubusercontent.com/vllm-project/vllm/v0.32.0/vllm/model_executor/models/registry.py\u003e (seen 2026-10-25)\n- SGLang v0.5.22 qwen3_coder parser (llama.py and the qwen3 reasoning parser at the same tag): \u003chttps://raw.githubusercontent.com/sgl-project/sglang/v0.5.22/python/sglang/srt/function_call/function_call_parser.py\u003e (seen 2026-10-25)\n- Transformers v5.20.2 llama: \u003chttps://raw.githubusercontent.com/huggingface/transformers/v5.20.2/src/transformers/models/llama/modeling_llama.py\u003e (seen 2026-10-25)\n- llama.cpp b11888 converter, LlamaForCausalLM mapped to llama: \u003chttps://raw.githubusercontent.com/ggml-org/llama.cpp/b11888/conversion/__init__.py\u003e (seen 2026-10-25)\n- Ollama search, no library entry for Nanbeige4.3: \u003chttps://ollama.com/search?q=nanbeige\u003e (seen 2026-10-25)\n- OpenRouter model list, no Nanbeige model: \u003chttps://openrouter.ai/api/v1/models\u003e (seen 2026-10-25)\n- BOSS Zhipin security response centre: \u003chttps://security.zhipin.com/\u003e (seen 2026-10-25)\n- Artificial Analysis leaderboard, Nanbeige4.1-3B only: \u003chttps://artificialanalysis.ai/models\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/nanbeige-4-3-33b/history.json\n\n- 2026-10-25: graded 61 C (methodology 0.6)\n\n## Who's behind it (provenance 53/100, checked 2026-10-25)\n\n| Check | Finding | Points |\n| --- | --- | --- |\n| Legal entity named | BOSS Zhipin (Kanzhun Limited), unconfirmed | 20/20 |\n| Domain age | zhipin.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 Apache-2.0 (Hub metadata only; the repository has no LICENSE file) licence stands in | 10/10 |\n| Privacy policy | nothing hosted, not scored | n/a |\n| Status page | not found | 0/10 |\n| Changelog | published | 10/10 |\n| security.txt | not found | 0/10 |\n\nThe card names the Nanbeige Team and nothing else. BOSS Zhipin is identified as the maker by the lab's Nanbeige4.2-3B report on arXiv (Nanbeige LLM Lab, Boss Zhipin), which links huggingface.co/Nanbeige, and by the chat template's default system prompt naming BOSS Zhipin as developer and trainer. The card's contact address is on kanzhun.com. Kanzhun Limited as BOSS Zhipin's operator wasn't confirmed first-hand, since ir.zhipin.com refused our reader. The domain's registration date wasn't checked.\n\nThe Hugging Face organisation Nanbeige (Nanbeige LLM Lab) isn't verified by the Hub. The arXiv report's link to it is the publisher's own documentation pointing there.\n\nNothing is hosted, so the licence stands in for terms of service and no privacy policy is scored.\n\nThe changelog is the repository's commit history.\n\nwww.zhipin.com/.well-known/security.txt answers 404 and kanzhun.com's redirects to the home page. BOSS Zhipin runs a security response centre at security.zhipin.com for reporting flaws in its products; nothing links it to the model.\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 Apache-2.0 (Hub metadata only; the repository has no LICENSE file) 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## 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- A plain LlamaForCausalLM with no repository code, so released vLLM, SGLang, Transformers and llama.cpp load it\n- Apache 2.0 by its metadata, ungated, safetensors only\n- 524,288 tokens of context set in config.json with no rope scaling\n- Thinking switches per request, and preserve_thinking keeps earlier reasoning for agent loops\n- Tool calls in an XML format the qwen3_coder parser reads, several per turn\n- A Nanbeige member answered four of the first ten discussions on release day\n\n## Weaknesses\n\n- No independent benchmark result yet; every agentic number on the card is Nanbeige's own run\n- BF16 only from Nanbeige, about 80 GB to load, and no Ollama or LM Studio build\n- No LICENSE file in either repository; Apache 2.0 rests on the metadata tag\n- No safety evaluation, agent-use guidance, training data description or knowledge cut-off\n- No hosted API or inference provider serves it\n\n\n\n## Before you call it (notes for agents)\n\n1. Serve with `vllm serve Nanbeige/Nanbeige4.3-33B --enable-auto-tool-choice --reasoning-parser qwen3 --tool-call-parser qwen3_coder`, as the card says, or SGLang with the same two parsers\n2. Keep `tool_call_format` at its default `xml`, since qwen3_coder parses that format; the `json` option writes Hermes-style JSON calls, which need a different parser such as vLLM's hermes\n3. Pass `chat_template_kwargs` `{\"preserve_thinking\": true}` for multi-turn tool use and coding, and `{\"enable_thinking\": false}` to skip reasoning, as the card recommends\n4. Use temperature 1.0, top_p 0.95 and up to 65,536 new tokens for agent work, and temperature 0.6 for reasoning and chat, per the card\n5. Send your own system prompt, since without one the template injects a Chinese identity line, or a Chinese tool-expert prompt when tools are passed\n6. Send one system message at the start only, as the template raises an error for a later one\n7. In plain Transformers pass `use_cache=True` to generate, since config.json and generation_config.json set it false; trust_remote_code isn't needed\n\n## Get started\n\nInstall:\n\n```bash\n# vLLM 0.32.0 or SGLang 0.5.22, as the card's commands (the card uses four GPUs with tensor parallelism)\nhf download Nanbeige/Nanbeige4.3-33B --local-dir Nanbeige4.3-33B\nvllm serve ./Nanbeige4.3-33B --served-model-name nanbeige --tensor-parallel-size 4 \\\n  --enable-auto-tool-choice --reasoning-parser qwen3 --tool-call-parser qwen3_coder\n# or\npython -m sglang.launch_server --model-path ./Nanbeige4.3-33B --tp-size 4 \\\n  --reasoning-parser qwen3 --tool-call-parser qwen3_coder\n```\n\nFirst request:\n\n```bash\ncurl http://localhost:8000/v1/chat/completions -H 'Content-Type: application/json' \\\n  -d '{\"model\": \"nanbeige\", \"messages\": [{\"role\": \"user\", \"content\": \"Which number is bigger, 9.11 or 9.8?\"}],\n       \"temperature\": 0.6, \"top_p\": 0.95, \"chat_template_kwargs\": {\"enable_thinking\": true}}'\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- Weights uploaded on 24 October 2026. Both repositories were created at 02:46 and 02:49 UTC that day, the BF16 shards went in at 11:57 UTC, and the full card was written on 25 October. The Hub doesn't record when the repositories turned public; the first discussion and the first copy by someone else both date from about 12:40 UTC on 25 October. On 25 October the Hub counted 344 likes (source: \u003chttps://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/commits/main\u003e)\n- Who makes it. The organisation is Nanbeige LLM Lab on Hugging Face (not verified by the Hub). The lab's Nanbeige4.2-3B report on arXiv gives its affiliation as Boss Zhipin and links the same organisation, this model's chat template tells the model it was developed and trained by BOSS Zhipin when no system prompt is given, and the card's contact address is on kanzhun.com, a site that loads BOSS Zhipin's assets (source: \u003chttps://arxiv.org/abs/2607.22083\u003e)\n- Nanbeige calls it the flagship teacher of the Nanbeige4.3 series, to be distilled into smaller models starting with Nanbeige4.3-3B, which isn't released (source: \u003chttps://huggingface.co/Nanbeige/Nanbeige4.3-33B\u003e)\n- Nanbeige's card claims the best agentic results among open models around 30B, ahead of Qwen3.8-27B, Muse-Glimmer-30B and Gemma4-31B on 15 of 17 agentic rows in its own runs, among them Terminal-Bench 2.1 80.9 with Terminus 2, DeepSWE 65.5, Toolathlon-Verified 70.4 and Tau3-Bench 48.8, with Qwen3.8-27B ahead on SWE-Bench Pro and GDPval. Some rows were run in Claude Code and mini-swe-agent harnesses. These are the publisher's numbers, and no independent result was found on 25 October 2026\n- Fine-tuned from Nanbeige4.3-33B-Base, published the same day in its own repository with the same 52-layer Llama shape, a 131,072-token context, a plain-text template and no discussions yet. Its card compares it only with Gemma4-31B-Base, in Nanbeige's own runs (source: \u003chttps://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base\u003e)\n- config.json was switched to a 1,048,576-token context with YaRN at 11:58 UTC on 24 October and back to 524,288 tokens with no scaling at 02:44 UTC on 25 October, both under the commit title Update config.json (source: \u003chttps://huggingface.co/api/models/Nanbeige/Nanbeige4.3-33B/commits/main\u003e)\n- Nanbeige4.2-3B, the previous release, is a looped transformer with modeling code in its repository and Nanbeige forks of vLLM, SGLang, llama.cpp and Ollama on its card. 4.3-33B is a plain LlamaForCausalLM with no repository code, so stock runtimes load it. Within a day others had posted more than 40 GGUF, MLX, FP8, NVFP4 and EXL3 copies, none from Nanbeige or a runtime's own library\n\n- #48 of 82 in Best open-weight models you download and run yourself for AI agents: https://www.anchorterminal.com/best/open-weights/index.md\n- All 3341 open weights comparisons: https://www.anchorterminal.com/compare/open-weights/index.md\n\n## Compare\n\n- [Agents-A1-4B vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/agents-a1-4b-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [Apertus v1.5 8B vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/apertus-v1-5-8b-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [Apodex-1.1-mini vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/apodex-1-1-mini-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [BigBang-v1 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/bigbang-v1-vs-nanbeige-4-3-33b.md): C 54.4 vs C 61\n- [Carbon-X vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/carbon-x-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [Cosmos-Reason2-2B vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/cosmos-reason2-2b-vs-nanbeige-4-3-33b.md): C 59.5 vs C 61\n- [Darwin-36B-Opus vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/darwin-36b-opus-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [DeepSeek-V3.2 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/deepseek-v3-2-vs-nanbeige-4-3-33b.md): C 58.6 vs C 61\n- [DeepSeek V4.1 Flash vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/deepseek-v4-1-flash-vs-nanbeige-4-3-33b.md): C 59 vs C 61\n- [DeepSeek-V4-Flash-0731 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/deepseek-v4-flash-0731-vs-nanbeige-4-3-33b.md): B 62.4 vs C 61\n- [DiffusionGemma 26B A4B IT vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/diffusiongemma-26b-a4b-it-vs-nanbeige-4-3-33b.md): D 50 vs C 61\n- [Gemma 4 12B IT vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/gemma-4-12b-it-vs-nanbeige-4-3-33b.md): BB 71.3 vs C 61\n- [Gemma 4 26B A4B IT vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/gemma-4-26b-a4b-it-vs-nanbeige-4-3-33b.md): BB 72.2 vs C 61\n- [Gemma 4 31B IT vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/gemma-4-31b-it-vs-nanbeige-4-3-33b.md): BB 74.1 vs C 61\n- [Gemma 4 E2B IT vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/gemma-4-e2b-it-vs-nanbeige-4-3-33b.md): C 60.4 vs C 61\n- [Gemma 4 E4B IT vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/gemma-4-e4b-it-vs-nanbeige-4-3-33b.md): B 63.2 vs C 61\n- [GLM-4.5-Air vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/glm-4-5-air-vs-nanbeige-4-3-33b.md): C 59.1 vs C 61\n- [GLM-4.7-Flash vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/glm-4-7-flash-vs-nanbeige-4-3-33b.md): B 63 vs C 61\n- [GLM-5.3-Flash vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/glm-5-3-flash-vs-nanbeige-4-3-33b.md): B 65.1 vs C 61\n- [GLM-5.3 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/glm-5-3-vs-nanbeige-4-3-33b.md): B 64.6 vs C 61\n- [gpt-oss-120b vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/gpt-oss-120b-vs-nanbeige-4-3-33b.md): BB 70.2 vs C 61\n- [gpt-oss-20b vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/gpt-oss-20b-vs-nanbeige-4-3-33b.md): B 69.9 vs C 61\n- [Granite 4.2 3B vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/granite-4-2-3b-vs-nanbeige-4-3-33b.md): C 59.8 vs C 61\n- [Hy3 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/hy3-vs-nanbeige-4-3-33b.md): B 62.6 vs C 61\n- [Inkling-Small vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/inkling-small-vs-nanbeige-4-3-33b.md): B 66.5 vs C 61\n- [Inkling vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/inkling-vs-nanbeige-4-3-33b.md): B 64.2 vs C 61\n- [InternVL3.5-GPT-OSS-20B-A4B-Preview vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/internvl3-5-gpt-oss-20b-a4b-preview-vs-nanbeige-4-3-33b.md): D 51.2 vs C 61\n- [Kimi K3 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/kimi-k3-vs-nanbeige-4-3-33b.md): B 62.4 vs C 61\n- [Kolibri-1 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/kolibri-1-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [Kronos vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/kronos-vs-nanbeige-4-3-33b.md): C 55.2 vs C 61\n- [Laguna M.1 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/laguna-m-1-vs-nanbeige-4-3-33b.md): C 55.5 vs C 61\n- [Laguna S 2.1 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/laguna-s-2-1-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [Laguna XS 2.1 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/laguna-xs-2-1-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [LFM2.5-2.6B vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/lfm2-5-2-6b-vs-nanbeige-4-3-33b.md): D 51.5 vs C 61\n- [Ling-3.0-tiny vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/ling-3-0-tiny-vs-nanbeige-4-3-33b.md): C 55 vs C 61\n- [MiMo-V2.6 Distill Qwen 9B vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/mimo-v2-6-distill-qwen-9b-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [MiniCPM-o 4.5 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/minicpm-o-4-5-vs-nanbeige-4-3-33b.md): C 61 vs C 61\n- [MiniCPM5-2B vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/minicpm5-2b-vs-nanbeige-4-3-33b.md): C 60.2 vs C 61\n- [MiniMax-M2.5 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/minimax-m2-5-vs-nanbeige-4-3-33b.md): C 57.5 vs C 61\n- [MiniMax-M2.7 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/minimax-m2-7-vs-nanbeige-4-3-33b.md): C 56.1 vs C 61\n- [MiniMax-M3.1-Flash vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/minimax-m3-1-flash-vs-nanbeige-4-3-33b.md): D 50.5 vs C 61\n- [Muse Glimmer 30B vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/muse-glimmer-30b-vs-nanbeige-4-3-33b.md): BB 70 vs C 61\n- [Nanbeige4.3-33B vs NVIDIA Nemotron 3.5 Bolt E4B 4GB](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-nemotron-3-5-bolt-e4b-4gb.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs Nemotron 3.5 Lightning 30B A3B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-nemotron-3-5-lightning-30b-a3b.md): C 61 vs B 64.2\n- [Nanbeige4.3-33B vs Nemotron 3 Nano 30B A3B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-nemotron-3-nano-30b-a3b.md): C 61 vs B 67.1\n- [Nanbeige4.3-33B vs Nemotron 3 Nano 4B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-nemotron-3-nano-4b.md): C 61 vs C 56.8\n- [Nanbeige4.3-33B vs Nemotron 3 Nano Omni 30B A3B Reasoning](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-nemotron-3-nano-omni-30b-a3b-reasoning.md): C 61 vs C 60.7\n- [Nanbeige4.3-33B vs Nemotron 3 Super 120B A12B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-nemotron-3-super-120b-a12b.md): C 61 vs B 66.2\n- [Nanbeige4.3-33B vs Nemotron 3 Ultra 550B A55B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-nemotron-3-ultra-550b-a55b.md): C 61 vs B 68.8\n- [Nanbeige4.3-33B vs occamy-1.0](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-occamy-1-0.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs Olmo 3 7B Think](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-olmo-3-7b-think.md): C 61 vs D 52.3\n- [Nanbeige4.3-33B vs Ornith-1.5-35B-A3B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-ornith-1-5-35b-a3b.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs Ornith-1.5-397B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-ornith-1-5-397b.md): C 61 vs C 59\n- [Nanbeige4.3-33B vs Ornith-1.5-9B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-ornith-1-5-9b.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs Qwen3-30B-A3B-Instruct-2507](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-30b-a3b-instruct-2507.md): C 61 vs C 56.6\n- [Nanbeige4.3-33B vs Qwen3-4B-Instruct-2507](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-4b-instruct-2507.md): C 61 vs C 55.7\n- [Nanbeige4.3-33B vs Qwen3.5-0.8B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-5-0-8b.md): C 61 vs D 46.7\n- [Nanbeige4.3-33B vs Qwen3.5-122B-A10B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-5-122b-a10b.md): C 61 vs B 63.6\n- [Nanbeige4.3-33B vs Qwen3.5-2B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-5-2b.md): C 61 vs D 50.5\n- [Nanbeige4.3-33B vs Qwen3.5-397B-A17B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-5-397b-a17b.md): C 61 vs B 64.6\n- [Nanbeige4.3-33B vs Qwen3.5-4B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-5-4b.md): C 61 vs C 61.2\n- [Nanbeige4.3-33B vs Qwen3.5-9B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-5-9b.md): C 61 vs C 60.5\n- [Nanbeige4.3-33B vs Qwen3.6-27B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-6-27b.md): C 61 vs B 67.2\n- [Nanbeige4.3-33B vs Qwen3.6-35B-A3B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-6-35b-a3b.md): C 61 vs B 63.2\n- [Nanbeige4.3-33B vs Qwen3.8-27B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-8-27b.md): C 61 vs B 68.6\n- [Nanbeige4.3-33B vs Qwen3.8-Flash-Next](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-8-flash-next.md): C 61 vs C 61.8\n- [Nanbeige4.3-33B vs Qwen3-Coder-30B-A3B-Instruct](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-coder-30b-a3b-instruct.md): C 61 vs C 55.7\n- [Nanbeige4.3-33B vs Qwen3-Coder-Next](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-coder-next.md): C 61 vs C 57.6\n- [Nanbeige4.3-33B vs Qwen3-Omni-30B-A3B-Instruct](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-omni-30b-a3b-instruct.md): C 61 vs E 41.8\n- [Nanbeige4.3-33B vs Qwen3-VL-2B-Instruct](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-vl-2b-instruct.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs Qwen3-VL-30B-A3B-Instruct](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-vl-30b-a3b-instruct.md): C 61 vs C 58\n- [Nanbeige4.3-33B vs Qwen3-VL-32B-Instruct](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-vl-32b-instruct.md): C 61 vs C 56.4\n- [Nanbeige4.3-33B vs Qwen3-VL-4B-Instruct](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-vl-4b-instruct.md): C 61 vs D 52.6\n- [Nanbeige4.3-33B vs Qwen3-VL-8B-Instruct](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwen3-vl-8b-instruct.md): C 61 vs C 54.9\n- [Nanbeige4.3-33B vs Qwopus3.8-27B-Flash-V2](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-qwopus3-8-27b-flash-v2.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs Recursive Flint](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-recursive-flint.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs SmolLM3-3B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-smollm3-3b.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs Spark-X2.6-4B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-spark-x2-6-4b.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs Swift 1.5 Qwen3.8-27B](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-swift-1-5-qwen3-8-27b.md): C 61 vs C 61\n- [Nanbeige4.3-33B vs Swift 1.5 Qwen3.8-Flash-Next](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-swift-1-5-qwen3-8-flash-next.md): C 61 vs D 52.3\n- [Nanbeige4.3-33B vs ThinkingCap (Qwen3.8 27B)](https://www.anchorterminal.com/compare/nanbeige-4-3-33b-vs-thinkingcap-qwen3-8-27b.md): C 61 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 on zhipin.com or one of its subdomains, or a page under huggingface.co/nanbeige. 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\": \"nanbeige-4-3-33b\", \"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/nanbeige-4-3-33b\"\u003e\u003cimg src=\"https://www.anchorterminal.com/badges/nanbeige-4-3-33b.svg\" alt=\"Nanbeige4.3-33B on Anchor Terminal\" height=\"20\"\u003e\u003c/a\u003e\n```\n\nMarkdown badge, for a README:\n\n```markdown\n[![Nanbeige4.3-33B on Anchor Terminal](https://www.anchorterminal.com/badges/nanbeige-4-3-33b.svg)](https://www.anchorterminal.com/tools/nanbeige-4-3-33b)\n```\n\nPlain link:\n\n```html\n\u003ca href=\"https://www.anchorterminal.com/tools/nanbeige-4-3-33b\"\u003eNanbeige4.3-33B 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 Nanbeige4.3-33B 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/nanbeige-4-3-33b-dark.png\n- Light: https://www.anchorterminal.com/assets/share/nanbeige-4-3-33b-light.png\n- Create a video and download its frames: https://www.anchorterminal.com/assets/listing-video/index.html?listing=nanbeige-4-3-33b\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": "2026-10-26",
    "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": "Nanbeige4.3-33B",
        "url": ""
      }
    ],
    "description": "Nanbeige4.3-33B is the Nanbeige LLM Lab's open-weight dense model for agentic coding and office work, 33B parameters on the Llama architecture with a 512K-token context, thinking that switches per request and tool calls in an XML format.",
    "facts": [
      "#48 in Open weights",
      "None auth",
      "0 desk reviews"
    ],
    "h1": "Nanbeige4.3-33B",
    "image": "https://www.anchorterminal.com/assets/og/tools-nanbeige-4-3-33b.png",
    "path": "/tools/nanbeige-4-3-33b",
    "published": "2026-10-01",
    "section": "tools",
    "title": "Nanbeige4.3-33B review (2026): pricing, alternatives and grade C",
    "toc": null,
    "updated": "2026-10-25",
    "url": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b"
  },
  "tokens": {
    "markdown": 9150,
    "slim": 2030
  },
  "version": 1
}