{ "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": "2026-11-11", "license": "CC-BY-4.0", "method": "https://www.anchorterminal.com/benchmark/", "methodology": "0.7", "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" }, "tool": { "slug": "nanbeige-4-2-3b", "name": "Nanbeige4.2-3B", "vendor": "Nanbeige LLM Lab (BOSS Zhipin)", "vendorUrl": "https://huggingface.co/Nanbeige", "kind": "weights", "category": "open-weights", "summary": "Nanbeige4.2-3B is a small agent model from BOSS Zhipin's Nanbeige lab, 4.2B parameters with layers reused in a loop, with tool calling, a thinking switch and a 256K context. Apache 2.0. Released SGLang and llama.cpp run it.", "url": "https://www.anchorterminal.com/tools/nanbeige-4-2-3b", "markdownUrl": "https://www.anchorterminal.com/tools/nanbeige-4-2-3b.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/nanbeige-4-2-3b.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/nanbeige-4-2-3b.json", "dns": "nanbeige-4-2-3b.grade.anchorterminal.com", "repo": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B", "license": "Apache-2.0 (Hub metadata only; the repository has no LICENSE file)", "transports": [], "packages": [ { "registry": "huggingface", "name": "Nanbeige/Nanbeige4.2-3B" }, { "registry": "huggingface", "name": "Nanbeige/Nanbeige4.2-3B-FP8" }, { "registry": "huggingface", "name": "Nanbeige/Nanbeige4.2-3B-GPTQ-Int8" } ], "auth": "none", "authNotes": "The weights download from Hugging Face with no account, token or terms to accept. None of the repositories is gated (checked 2026-11-10).", "pricing": "free", "pricingNotes": "Free to download and use under Apache 2.0, per the card's metadata. Running it yourself costs little: 5.2 GB in FP8 or 8-bit from Nanbeige, 8.3 GB in BF16, and 2.7 GB for bartowski's Q4_K_M conversion (checked 2026-11-10).", "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 2026-11-10).", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-11-10" }, "docsUrl": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B", "capabilities": [ "model.open-weights" ], "tags": [ "open-weights", "model", "small", "looped-transformer", "reasoning", "tool-calling", "agentic", "long-context", "apache-2.0", "safetensors", "fp8", "custom-code", "self-hosted" ], "lastRelease": "2026-09-11", "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": 1518, "categoryRank": 62, "methodology": "0.7", "run": "2026-10-01", "scores": { "interface": 85, "reach": 85, "risk": 55, "upkeep": 72 }, "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": 85, "points": 32.69, "reason": "Memory 30, the smallest build from Nanbeige is the FP8 repository (the 8-bit GPTQ one is the same size), 5.19 GB, and 5.19 GB × 1.2 = 6.2 GB, within 16 GB. bartowski's GGUF isn't read for this line. Runtimes 15, Released SGLang 0.5.23 has the model (srt/models/nanbeige.py) with its parsers, llama.cpp b12242 has the nanbeige architecture, and vLLM 0.33.0 names NanbeigeForCausalLM in its registry, served through its Transformers backend. Builds 15, Quantised builds from Nanbeige at two precisions, FP8 and 8-bit GPTQ, each in its own repository. Terms 15, Apache 2.0 in the card's metadata, an OSI licence, with no use policy or other terms on the card. Download 10, the Hub API reports gated false for the model and its FP8 and 8-bit repositories on 10 November 2026, so the files download with no account and nothing to accept. Hosted 0, Hugging Face maps no inference provider to it, and no Nanbeige or BOSS Zhipin API for it was found. 30 + 15 + 15 + 15 + 10 + 0 = 85.", "checked": "2026-11-10" }, { "key": "risk", "name": "Risk", "weight": 20, "effectiveWeight": 30.77, "score": 55, "points": 16.92, "reason": "File safety 15, safetensors only, no .bin, .pt or .pth files (25). Provenance 15, the files sit in the Nanbeige organisation (Nanbeige LLM Lab), which the lab's arXiv report for this model links, so the publisher's own account, although the Hub hasn't verified it (10). Safety evaluation 0, no safety results on the card. Agent guidance 5, General cautions: the limitations section says outputs may be inaccurate, biased or harmful. Licence risk 15, Apache 2.0 has no use policy, no threshold and no clause letting the licensor change or revoke it. Security programme 5, BOSS Zhipin runs a security response centre for reporting flaws in its products, as read for the Nanbeige4.3-33B listing (5), though nothing links it to the model. 15 + 15 + 0 + 5 + 15 + 5 = 55.", "checked": "2026-11-10" }, { "key": "interface", "name": "Agent interface", "weight": 10, "effectiveWeight": 15.38, "score": 85, "points": 13.08, "reason": "Chat template 20, A Jinja chat template in tokenizer_config.json (7,961 characters), with enable_thinking, preserve_thinking, tools and tool_call_format switches. Tool-calling format 25, the card says passing tools switches on the tool template and recommends tool_call_format xml, with json for compatibility, and its SGLang and vLLM commands name the nanbeige tool-call parser, which released SGLang 0.5.23 has. Trained tool calling 20, the card describes fine-tuning on tool-use environments and agent scaffolds and reinforcement learning on agent tasks, and the template tells the model it may call one or more functions and renders every call in a turn. Structured output 0, Neither the card nor the files mention JSON or JSON Schema output, beyond json as a tool-call format. Reasoning control 10, the card documents enable_thinking, on by default and False per request, and preserve_thinking for earlier turns, and names the nanbeige reasoning parser, which returns the reasoning apart from the answer. Context 10, 262,144 tokens, which the card says the model supports and config.json gives as max_position_embeddings. 20 + 25 + 20 + 0 + 10 + 10 = 85.", "checked": "2026-11-10" }, { "key": "upkeep", "name": "Documentation and upkeep", "weight": 10, "effectiveWeight": 15.38, "score": 72, "points": 11.08, "reason": "Model card 17, Intended use, limitations, the architecture, the parameters (3B outside the embeddings), the context and the languages (in the metadata) are on the card. Training data 5, A general description of how the fine-tuning and reinforcement learning data was built and filtered, with no sources, sizes or cut-off. Last change 15, the last commit, on 11 September 2026, updated modeling_nanbeige.py, 60 days before this check and within 90. Responsiveness 20, Of the 34 most recent discussions that aren't pull requests (21 July to 10 September 2026), 20 have a reply from a Nanbeige member, which is most. Version policy 5, Nanbeige's earlier models (Nanbeige4-3B, 4.1-3B) stay in their own repositories, as the Base, FP8, 8-bit and DSpark builds of this one do. File health 10, Config, tokenizer config, template and card agree on the architecture and the 262,144 tokens. 17 + 5 + 15 + 20 + 5 + 10 = 72.", "checked": "2026-11-10" } ], "assessment": { "date": "2026-11-10", "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": "Chat template 20, A Jinja chat template in tokenizer_config.json (7,961 characters), with enable_thinking, preserve_thinking, tools and tool_call_format switches. Tool-calling format 25, the card says passing tools switches on the tool template and recommends tool_call_format xml, with json for compatibility, and its SGLang and vLLM commands name the nanbeige tool-call parser, which released SGLang 0.5.23 has. Trained tool calling 20, the card describes fine-tuning on tool-use environments and agent scaffolds and reinforcement learning on agent tasks, and the template tells the model it may call one or more functions and renders every call in a turn. Structured output 0, Neither the card nor the files mention JSON or JSON Schema output, beyond json as a tool-call format. Reasoning control 10, the card documents enable_thinking, on by default and False per request, and preserve_thinking for earlier turns, and names the nanbeige reasoning parser, which returns the reasoning apart from the answer. Context 10, 262,144 tokens, which the card says the model supports and config.json gives as max_position_embeddings. 20 + 25 + 20 + 0 + 10 + 10 = 85.", "reach": "Memory 30, the smallest build from Nanbeige is the FP8 repository (the 8-bit GPTQ one is the same size), 5.19 GB, and 5.19 GB × 1.2 = 6.2 GB, within 16 GB. bartowski's GGUF isn't read for this line. Runtimes 15, Released SGLang 0.5.23 has the model (srt/models/nanbeige.py) with its parsers, llama.cpp b12242 has the nanbeige architecture, and vLLM 0.33.0 names NanbeigeForCausalLM in its registry, served through its Transformers backend. Builds 15, Quantised builds from Nanbeige at two precisions, FP8 and 8-bit GPTQ, each in its own repository. Terms 15, Apache 2.0 in the card's metadata, an OSI licence, with no use policy or other terms on the card. Download 10, the Hub API reports gated false for the model and its FP8 and 8-bit repositories on 10 November 2026, so the files download with no account and nothing to accept. Hosted 0, Hugging Face maps no inference provider to it, and no Nanbeige or BOSS Zhipin API for it was found. 30 + 15 + 15 + 15 + 10 + 0 = 85.", "risk": "File safety 15, safetensors only, no .bin, .pt or .pth files (25). Provenance 15, the files sit in the Nanbeige organisation (Nanbeige LLM Lab), which the lab's arXiv report for this model links, so the publisher's own account, although the Hub hasn't verified it (10). Safety evaluation 0, no safety results on the card. Agent guidance 5, General cautions: the limitations section says outputs may be inaccurate, biased or harmful. Licence risk 15, Apache 2.0 has no use policy, no threshold and no clause letting the licensor change or revoke it. Security programme 5, BOSS Zhipin runs a security response centre for reporting flaws in its products, as read for the Nanbeige4.3-33B listing (5), though nothing links it to the model. 15 + 15 + 0 + 5 + 15 + 5 = 55.", "upkeep": "Model card 17, Intended use, limitations, the architecture, the parameters (3B outside the embeddings), the context and the languages (in the metadata) are on the card. Training data 5, A general description of how the fine-tuning and reinforcement learning data was built and filtered, with no sources, sizes or cut-off. Last change 15, the last commit, on 11 September 2026, updated modeling_nanbeige.py, 60 days before this check and within 90. Responsiveness 20, Of the 34 most recent discussions that aren't pull requests (21 July to 10 September 2026), 20 have a reply from a Nanbeige member, which is most. Version policy 5, Nanbeige's earlier models (Nanbeige4-3B, 4.1-3B) stay in their own repositories, as the Base, FP8, 8-bit and DSpark builds of this one do. File health 10, Config, tokenizer config, template and card agree on the architecture and the 262,144 tokens. 17 + 5 + 15 + 20 + 5 + 10 = 72." }, "sources": [ { "what": "model card", "url": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B", "seen": "2026-11-10" }, { "what": "gating, dates, parameter count, downloads, likes, file list and hashes", "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.2-3B?blobs=true", "seen": "2026-11-10" }, { "what": "the FP8 repository", "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.2-3B-FP8?blobs=true", "seen": "2026-11-10" }, { "what": "the 8-bit GPTQ repository", "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.2-3B-GPTQ-Int8?blobs=true", "seen": "2026-11-10" }, { "what": "inference providers, none mapped", "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.2-3B?expand[]=inferenceProviderMapping", "seen": "2026-11-10" }, { "what": "commit history", "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.2-3B/commits/main", "seen": "2026-11-10" }, { "what": "architecture, auto_map and positions", "url": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B/raw/main/config.json", "seen": "2026-11-10" }, { "what": "tokenizer config, with the chat template", "url": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B/raw/main/tokenizer_config.json", "seen": "2026-11-10" }, { "what": "discussions, replies counted by script from authors, dates and member flags", "url": "https://huggingface.co/api/models/Nanbeige/Nanbeige4.2-3B/discussions", "seen": "2026-11-10" }, { "what": "bartowski's GGUF repository, a third party's conversion", "url": "https://huggingface.co/api/models/bartowski/Nanbeige_Nanbeige4.2-3B-GGUF?blobs=true", "seen": "2026-11-10" }, { "what": "SGLang v0.5.23, srt/models/nanbeige.py and the parser names", "url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.23/python/sglang/srt/models/nanbeige.py", "seen": "2026-11-10" }, { "what": "vLLM v0.33.0 registry, NanbeigeForCausalLM on the Transformers backend", "url": "https://raw.githubusercontent.com/vllm-project/vllm/v0.33.0/vllm/model_executor/models/registry.py", "seen": "2026-11-10" }, { "what": "llama.cpp b12242, the nanbeige architecture (shallow clone)", "url": "https://raw.githubusercontent.com/ggml-org/llama.cpp/b12242/src/llama-arch.cpp", "seen": "2026-11-10" }, { "what": "our Nanbeige4.3-33B listing, for company facts", "url": "https://www.anchorterminal.com/tools/nanbeige-4-3-33b", "seen": "2026-11-10" } ], "openQuestions": [ "vLLM is credited because its registry names the architecture, though it serves it through the Transformers backend and the card's vLLM parser flags come from Nanbeige's fork. A stricter reading gives Runtimes 10", "File safety takes the 10 off for the Transformers path. Someone who only uses SGLang or llama.cpp runs no repository code", "The card cites Artificial Analysis's small-model leaderboard, which an independent read would turn into a Capability score", "Capability is pending because no independent results were looked up on the day", "unchecked on 10 November: the technical report (Nanbeige42_report.pdf, arXiv 2607.22083), the repository's Python files, Nanbeige's forks, Ollama's pinned llama.cpp, and LM Studio's catalogue" ] }, "negative": 0, "verdict": "A small model built for agent work, with tool calling it was trained for, a thinking switch and a 256K context in about 5 GB, under Apache 2.0, from a lab that answers most questions. Its architecture is its own, so check your runtime: SGLang and llama.cpp have it, Transformers needs the repository's code, and Nanbeige ships no GGUF.", "bestFor": "Local agents that need tool calling and a long context on a laptop or a small GPU", "strengths": [ "Trained for multi-step tool use, with parsers in released SGLang", "A 256K context in a model that fits 8 GB", "Apache 2.0, ungated", "FP8 and 8-bit builds from Nanbeige", "Nanbeige replied in most of the recent discussions" ], "weaknesses": [ "Its own architecture: Transformers loads it only with the repository's code", "No GGUF from Nanbeige, so llama.cpp users rely on a third party's conversion", "The card's serve instructions still point at Nanbeige's forks", "No safety results on the card", "Training data described only in general terms", "Not changed since September" ], "agentNotes": [ "Serve it with SGLang 0.5.23 or later and pass `--tool-call-parser nanbeige --reasoning-parser nanbeige`", "Pass `tools` in the request. The card recommends `tool_call_format=\"xml\"`", "Set `preserve_thinking` true for multi-turn tool use and coding, false for plain chat", "Use temperature 1.0 for agent and tool work and 0.6 for reasoning and chat, per the card", "For llama.cpp use b12242 or later. The GGUF files on the Hub are bartowski's, not Nanbeige's", "If you load it with Transformers, read modeling_nanbeige.py first: trust_remote_code runs it" ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "low", "grade": "C", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 61 } ], "editorialScores": { "interface": 85, "reach": 85, "risk": 55, "upkeep": 72 }, "provenanceScore": 58, "checked": { "categories": { "interface": "2026-11-10", "reach": "2026-11-10", "risk": "2026-11-10", "upkeep": "2026-11-10" }, "last": "2026-11-10", "oldest": "2026-11-10", "stale": false, "staleAfterDays": 90 }, "scoreHistory": [ { "date": "2026-11-10", "grade": "C", "kind": "listed", "method": "0.7", "score": 61 } ] }, "connect": { "install": "pip install \"sglang\u003e=0.5.23\"\npython -m sglang.launch_server --model-path Nanbeige/Nanbeige4.2-3B --reasoning-parser nanbeige --tool-call-parser nanbeige", "http": "# SGLang serves an OpenAI-compatible API at http://localhost:30000/v1. Pass tools in the request, and\n# chat_template_kwargs {\"enable_thinking\": false} for a direct answer" }, "letme": { "capability": "https://letme.dev/model.open-weights", "tool": "https://letme.dev/nanbeige-4-2-3b" }, "sameCompany": [ "nanbeige-4-3-33b" ], "weights": { "trainedBy": "Nanbeige LLM Lab (BOSS Zhipin)", "builtOn": "Nanbeige4.2-3B-Base, per the card and the metadata", "params": "4.2B", "context": 262144, "goodFor": "A small local agent for tool use, coding and office tasks, with thinking you can turn off and a long context.", "runtimes": [ "llama.cpp", "vLLM", "SGLang" ], "builds": [ { "name": "Nanbeige4.2-3B-FP8", "url": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B-FP8", "who": "publisher", "precision": "fp8", "sizeGB": 5.19, "memoryGB": 8, "format": "safetensors", "runsOn": [ "vLLM", "SGLang" ], "note": "Nanbeige's FP8 build" }, { "name": "Nanbeige4.2-3B-GPTQ-Int8", "url": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B-GPTQ-Int8", "who": "publisher", "precision": "int8", "sizeGB": 5.19, "memoryGB": 8, "format": "safetensors", "runsOn": [ "vLLM", "SGLang" ], "note": "Nanbeige's 8-bit GPTQ build" }, { "name": "bartowski Nanbeige_Nanbeige4.2-3B-GGUF (Q4_K_M)", "url": "https://huggingface.co/bartowski/Nanbeige_Nanbeige4.2-3B-GGUF", "who": "bartowski", "precision": "q4_k_m", "sizeGB": 2.68, "memoryGB": 8, "format": "gguf", "runsOn": [ "llama.cpp" ], "note": "From bartowski, a third party's conversion and not Nanbeige's. Its base_model tag names Nanbeige/Nanbeige4.2-3B, from the Hub's model information on 10 November 2026" }, { "name": "bartowski Nanbeige_Nanbeige4.2-3B-GGUF (Q8_0)", "url": "https://huggingface.co/bartowski/Nanbeige_Nanbeige4.2-3B-GGUF", "who": "bartowski", "precision": "q8_0", "sizeGB": 4.43, "memoryGB": 8, "format": "gguf", "runsOn": [ "llama.cpp" ], "note": "From bartowski, a third party's conversion and not Nanbeige's. From the Hub's model information on 10 November 2026" }, { "name": "Nanbeige4.2-3B (BF16 safetensors)", "url": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B", "who": "publisher", "precision": "bf16", "sizeGB": 8.34, "memoryGB": 12, "format": "safetensors", "runsOn": [ "vLLM", "SGLang" ], "note": "Two shards" } ], "run": "python -m sglang.launch_server --model-path Nanbeige/Nanbeige4.2-3B --reasoning-parser nanbeige --tool-call-parser nanbeige", "runBuild": "Nanbeige4.2-3B (BF16 safetensors)", "checked": "2026-11-10", "minMemoryGB": 8 }, "family": { "lab": "Nanbeige LLM Lab (BOSS Zhipin)", "family": "Nanbeige4.2", "url": "https://www.anchorterminal.com/open-weights/#nanbeige", "others": [ { "name": "Nanbeige4.2-3B-Base", "url": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B-Base" } ], "olderCount": 7 }, "notable": [ "108,446 downloads in the last 30 days and 859 likes on Hugging Face on 10 November 2026 (https://huggingface.co/api/models/Nanbeige/Nanbeige4.2-3B)", "A looped transformer: the layers are reused to add depth without adding parameters, which the card puts at 3B outside the embeddings; the files hold 4.17B (https://huggingface.co/Nanbeige/Nanbeige4.2-3B)", "The card's own instructions for SGLang, vLLM and llama.cpp clone Nanbeige's forks. Since then released SGLang 0.5.23 and llama.cpp b12242 carry the architecture themselves (https://huggingface.co/Nanbeige/Nanbeige4.2-3B)", "The card says LM Studio's bundled llama-server doesn't support the architecture yet, and gives a workaround (https://huggingface.co/Nanbeige/Nanbeige4.2-3B)", "The card says the model was trained on tool-use environments with filtering at the trajectory and turn level, then with reinforcement learning (https://huggingface.co/Nanbeige/Nanbeige4.2-3B)", "A Nanbeige member replied in 20 of the 34 most recent discussions (https://huggingface.co/api/models/Nanbeige/Nanbeige4.2-3B/discussions)", "Nanbeige says its next model, Nanbeige4.5, is in training for release later in 2026 (https://huggingface.co/Nanbeige/Nanbeige4.2-3B)" ], "area": "models", "details": [ { "label": "Surface graded", "value": "The open weights at huggingface.co/Nanbeige/Nanbeige4.2-3B and the FP8, GPTQ-Int8 and DSpark repositories beside it, read from the model card, the files and their metadata, the config, the tokenizer config with its chat template, the commit history and the runtimes' own code at their latest releases on 10 November 2026. Nanbeige's company facts come from our Nanbeige4.3-33B listing. The technical report in the repository, the repository's Python files and Nanbeige's forks weren't read. Nothing was downloaded or run" }, { "label": "Model", "value": "4,169,800,704 parameters in the files; the card says 3B outside the embeddings. NanbeigeForCausalLM, a looped transformer that reuses its layers, built on Nanbeige4.2-3B-Base with supervised fine-tuning on agent trajectories and reinforcement learning" }, { "label": "Files", "value": "Two safetensors shards in BF16 (8.34 GB), config and tokenizer files with the chat template in tokenizer_config.json, two Python modules (configuration_nanbeige.py, modeling_nanbeige.py), the technical report as a PDF and evaluation result files. No pickle files. Hugging Face lists a SHA-256 for the large files" }, { "label": "Licence", "value": "Apache 2.0 in the metadata of every repository. There is no LICENSE file and the card's text names no licence" }, { "label": "Runtimes", "value": "The card documents Transformers 4.45.1 with trust_remote_code, and SGLang, vLLM, llama.cpp and Ollama from Nanbeige's own forks. Released SGLang 0.5.23 has the model and its tool and reasoning parsers, llama.cpp b12242 has the architecture, and vLLM 0.33.0 names it in its registry and runs it through its Transformers backend. Transformers 5.22.2 has no native model" }, { "label": "Context", "value": "262,144 tokens, per the card and config.json" }, { "label": "Tools and reasoning", "value": "Passing tools switches on the tool template, with tool_call_format xml recommended and json supported. enable_thinking is on by default and off per request, and preserve_thinking keeps earlier reasoning for multi-turn tool use. The card's serve commands name the nanbeige tool-call and reasoning parsers" }, { "label": "Hardware", "value": "No memory figure on the card. 8.34 GB in BF16 and 5.19 GB in FP8 or 8-bit. The card presents it as a local personal assistant" }, { "label": "Training data", "value": "A general description on the card: supervised fine-tuning across real and synthesised tool-use environments with filtering by test cases and rubrics, then reinforcement learning with outcome and process rewards. No sources, sizes or cut-off. The technical report wasn't read" } ], "provenance": { "legalEntity": "BOSS Zhipin (Kanzhun Limited), unconfirmed", "domain": "zhipin.com", "domainRegistered": "", "endpointOnVendorDomain": null, "terms": "", "privacy": "", "statusPage": "", "changelog": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B/commits/main", "securityTxt": "none", "checked": "2026-11-10", "notes": [ "The publisher's identity, domain and security facts are carried from our Nanbeige4.3-33B listing: the lab's report for this model names Nanbeige LLM Lab, BOSS Zhipin, and the card's contact address is on kanzhun.com.", "Nothing is hosted under this listing, so Apache 2.0 stands in for terms of service and no privacy policy is scored.", "The changelog is the commit history, 29 commits from 21 July to 11 September 2026." ], "score": 58, "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" }, { "check": "People named", "value": "1 named with their role, none with a public profile", "points": 7, "max": 10, "state": "part" }, { "check": "Support route", "value": "help pages", "points": 5, "max": 5, "state": "ok" } ] }, "humans": { "slug": "nanbeige-4-2-3b", "checked": "2026-11-10", "company": "Nanbeige LLM Lab (Kanzhun, BOSS Zhipin)", "people": [ { "name": "Peng Zhao", "role": "Founder, chairman and CEO, Kanzhun (BOSS Zhipin)", "founder": true, "bio": "Peng Zhao founded Kanzhun in Beijing in January 2014 as a company review website; it later launched the recruitment app BOSS Zhipin. He is the company's chairman and chief executive.", "bioSources": [ "https://en.wikipedia.org/wiki/Kanzhun" ], "page": "/people/peng-zhao/" } ], "peopleNote": "The Nanbeige model card credits only the Nanbeige Team and gives a Kanzhun contact address. Kanzhun's founder and chief executive is from Wikipedia; Kanzhun's investor relations site and SEC EDGAR refused to serve their robots.txt, so their management pages weren't read. Roster may be incomplete.", "companySocial": {}, "support": { "urls": [ "https://huggingface.co/Nanbeige/Nanbeige4.2-3B/discussions" ] }, "sources": [ "https://huggingface.co/Nanbeige/Nanbeige4.2-3B", "https://huggingface.co/api/organizations/Nanbeige/overview", "https://en.wikipedia.org/wiki/Kanzhun" ], "ownership": { "status": "subsidiary", "parent": "Kanzhun Limited", "checked": "2026-10-28", "sources": [ "https://huggingface.co/Nanbeige", "https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0825/2026082501403.pdf" ] } }, "pageJsonUrl": "https://www.anchorterminal.com/tools/nanbeige-4-2-3b.json" } }