{ "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": "zaya1-8b", "name": "ZAYA1-8B", "vendor": "Zyphra", "vendorUrl": "https://www.zyphra.com", "kind": "weights", "category": "open-weights", "summary": "ZAYA1-8B is Zyphra's open-weight reasoning model, a mixture of experts of 8.4B parameters with 760M active, trained from scratch for maths and coding. Apache 2.0. Zyphra's 4-bit build is 5.9 GB. Released SGLang and Transformers run it.", "url": "https://www.anchorterminal.com/tools/zaya1-8b", "markdownUrl": "https://www.anchorterminal.com/tools/zaya1-8b.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/zaya1-8b.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/zaya1-8b.json", "dns": "zaya1-8b.grade.anchorterminal.com", "repo": "https://huggingface.co/Zyphra/ZAYA1-8B", "license": "Apache-2.0", "transports": [], "packages": [ { "registry": "huggingface", "name": "Zyphra/ZAYA1-8B" }, { "registry": "huggingface", "name": "Zyphra/ZAYA1-8B-FP8-Experts" }, { "registry": "huggingface", "name": "Zyphra/ZAYA1-8B-MXFP4-Experts" } ], "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 main repository's metadata. Running it yourself costs little: 5.9 GB with the experts at 4 bits (MXFP4), 9.6 GB in FP8 or 17.7 GB in BF16 (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/Zyphra/ZAYA1-8B", "capabilities": [ "model.open-weights" ], "tags": [ "open-weights", "model", "moe", "small", "reasoning", "maths", "coding", "tool-calling", "apache-2.0", "safetensors", "fp8", "mxfp4", "self-hosted" ], "lastRelease": "2026-07-02", "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": 67, "methodology": "0.7", "run": "2026-10-01", "scores": { "interface": 60, "reach": 80, "risk": 55, "upkeep": 50 }, "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": 80, "points": 30.77, "reason": "Memory 30, the smallest build from Zyphra is the MXFP4-Experts repository, 5.85 GB, and 5.85 GB × 1.2 = 7.0 GB, within 16 GB. Runtimes 10, Released Transformers 5.22.2 has the zaya model and SGLang 0.5.23 has srt/models/zaya.py. Builds 15, Quantised builds from Zyphra at two precisions, FP8 and MXFP4 for the experts, each in its own repository. Terms 15, Apache 2.0 in the card's metadata, an OSI licence with nothing added. Download 10, the Hub API reports gated false for the model and its builds 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 the card names no API. Zyphra's site lists a cloud product for its open models, which wasn't checked for this model. 30 + 10 + 15 + 15 + 10 + 0 = 80.", "checked": "2026-11-10" }, { "key": "risk", "name": "Risk", "weight": 20, "effectiveWeight": 30.77, "score": 55, "points": 16.92, "reason": "File safety 25, safetensors only, no .bin, .pt or .pth files and no Python in the repository. Provenance 15, the files sit in the Zyphra organisation, which the card ties to zyphra.com's blog post for the model (10). Safety evaluation 0, no safety results on the card. Agent guidance 0, the card has no limitations or safety section, and nothing on prompt injection, instruction priority or what the model refuses. Licence risk 15, Apache 2.0 has no use policy, threshold or change clause. Security programme 0, no way to report a flaw on the card, and www.zyphra.com has no security.txt. 25 + 15 + 0 + 0 + 15 + 0 = 55.", "checked": "2026-11-10" }, { "key": "interface", "name": "Agent interface", "weight": 10, "effectiveWeight": 15.38, "score": 60, "points": 9.23, "reason": "Chat template 20, chat_template.jinja in the repository (10,700 characters). Tool-calling format 10, only the template and a fork say it. Trained tool calling 10, the template takes tools and renders calls, and the card recommends settings for agent use, but it doesn't say the model was trained for tool calling or for several calls in a turn, so the lower band. Structured output 0, Nothing on the card about JSON or schema-constrained output. Reasoning control 10, the template has an enable_thinking switch, and the card names the qwen3 reasoning parser, which returns the reasoning apart from the answer. Context 10, 131,072 positions in config.json, for an architecture Zyphra trained end to end, so the config is the model's own. 20 + 10 + 10 + 0 + 10 + 10 = 60.", "checked": "2026-11-10" }, { "key": "upkeep", "name": "Documentation and upkeep", "weight": 10, "effectiveWeight": 15.38, "score": 50, "points": 7.69, "reason": "Model card 5, Intended use and the parameter counts are on the card. Training data 0, Nothing on the card about the training data. Last change 10, the main repository last changed on 26 June 2026 (the checkpoint reshaped for Transformers), 137 days before this check, and the quantised repositories on 2 July, 131 days. Responsiveness 20, Of the 16 most recent discussions that aren't pull requests (7 May to 14 August 2026), 10 have a reply from a Zyphra member, which is most. Version policy 5, the original checkpoint is kept as ZAYA1-8B-legacy and the card says so at the top, without a date. File health 10, Config, template and card agree on the architecture after the reshape. 5 + 0 + 10 + 20 + 5 + 10 = 50.", "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, chat_template.jinja in the repository (10,700 characters). Tool-calling format 10, only the template and a fork say it. Trained tool calling 10, the template takes tools and renders calls, and the card recommends settings for agent use, but it doesn't say the model was trained for tool calling or for several calls in a turn, so the lower band. Structured output 0, Nothing on the card about JSON or schema-constrained output. Reasoning control 10, the template has an enable_thinking switch, and the card names the qwen3 reasoning parser, which returns the reasoning apart from the answer. Context 10, 131,072 positions in config.json, for an architecture Zyphra trained end to end, so the config is the model's own. 20 + 10 + 10 + 0 + 10 + 10 = 60.", "reach": "Memory 30, the smallest build from Zyphra is the MXFP4-Experts repository, 5.85 GB, and 5.85 GB × 1.2 = 7.0 GB, within 16 GB. Runtimes 10, Released Transformers 5.22.2 has the zaya model and SGLang 0.5.23 has srt/models/zaya.py. Builds 15, Quantised builds from Zyphra at two precisions, FP8 and MXFP4 for the experts, each in its own repository. Terms 15, Apache 2.0 in the card's metadata, an OSI licence with nothing added. Download 10, the Hub API reports gated false for the model and its builds 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 the card names no API. Zyphra's site lists a cloud product for its open models, which wasn't checked for this model. 30 + 10 + 15 + 15 + 10 + 0 = 80.", "risk": "File safety 25, safetensors only, no .bin, .pt or .pth files and no Python in the repository. Provenance 15, the files sit in the Zyphra organisation, which the card ties to zyphra.com's blog post for the model (10). Safety evaluation 0, no safety results on the card. Agent guidance 0, the card has no limitations or safety section, and nothing on prompt injection, instruction priority or what the model refuses. Licence risk 15, Apache 2.0 has no use policy, threshold or change clause. Security programme 0, no way to report a flaw on the card, and www.zyphra.com has no security.txt. 25 + 15 + 0 + 0 + 15 + 0 = 55.", "upkeep": "Model card 5, Intended use and the parameter counts are on the card. Training data 0, Nothing on the card about the training data. Last change 10, the main repository last changed on 26 June 2026 (the checkpoint reshaped for Transformers), 137 days before this check, and the quantised repositories on 2 July, 131 days. Responsiveness 20, Of the 16 most recent discussions that aren't pull requests (7 May to 14 August 2026), 10 have a reply from a Zyphra member, which is most. Version policy 5, the original checkpoint is kept as ZAYA1-8B-legacy and the card says so at the top, without a date. File health 10, Config, template and card agree on the architecture after the reshape. 5 + 0 + 10 + 20 + 5 + 10 = 50." }, "sources": [ { "what": "model card", "url": "https://huggingface.co/Zyphra/ZAYA1-8B", "seen": "2026-11-10" }, { "what": "gating, dates, parameter count, downloads, likes, file list and hashes", "url": "https://huggingface.co/api/models/Zyphra/ZAYA1-8B?blobs=true", "seen": "2026-11-10" }, { "what": "the FP8-Experts repository", "url": "https://huggingface.co/api/models/Zyphra/ZAYA1-8B-FP8-Experts?blobs=true", "seen": "2026-11-10" }, { "what": "the MXFP4-Experts repository", "url": "https://huggingface.co/api/models/Zyphra/ZAYA1-8B-MXFP4-Experts?blobs=true", "seen": "2026-11-10" }, { "what": "inference providers, none mapped", "url": "https://huggingface.co/api/models/Zyphra/ZAYA1-8B?expand[]=inferenceProviderMapping", "seen": "2026-11-10" }, { "what": "commit history", "url": "https://huggingface.co/api/models/Zyphra/ZAYA1-8B/commits/main", "seen": "2026-11-10" }, { "what": "architecture and positions", "url": "https://huggingface.co/Zyphra/ZAYA1-8B/raw/main/config.json", "seen": "2026-11-10" }, { "what": "chat template", "url": "https://huggingface.co/Zyphra/ZAYA1-8B/raw/main/chat_template.jinja", "seen": "2026-11-10" }, { "what": "discussions, replies counted by script from authors, dates and member flags", "url": "https://huggingface.co/api/models/Zyphra/ZAYA1-8B/discussions", "seen": "2026-11-10" }, { "what": "the organisation's profile on the Hub, not marked verified", "url": "https://huggingface.co/api/organizations/Zyphra/overview", "seen": "2026-11-10" }, { "what": "robots.txt for www.zyphra.com", "url": "https://www.zyphra.com/robots.txt", "seen": "2026-11-10" }, { "what": "Zyphra's about page", "url": "https://www.zyphra.com/about", "seen": "2026-11-10" }, { "what": "Zyphra's home page", "url": "https://www.zyphra.com/", "seen": "2026-11-10" }, { "what": "security.txt, 404", "url": "https://www.zyphra.com/.well-known/security.txt", "seen": "2026-11-10" }, { "what": "SGLang v0.5.23, srt/models/zaya.py", "url": "https://raw.githubusercontent.com/sgl-project/sglang/v0.5.23/python/sglang/srt/models/zaya.py", "seen": "2026-11-10" }, { "what": "Transformers v5.22.2 zaya", "url": "https://github.com/huggingface/transformers/tree/v5.22.2/src/transformers/models/zaya", "seen": "2026-11-10" }, { "what": "vLLM v0.33.0 registry and llama.cpp b12242, no zaya", "url": "https://raw.githubusercontent.com/vllm-project/vllm/v0.33.0/vllm/model_executor/models/registry.py", "seen": "2026-11-10" } ], "openQuestions": [ "Context is 10 on config.json alone, since Zyphra trained the architecture itself. The card states no length", "Which released runtime loads the FP8 and MXFP4 expert builds wasn't established, so the build that fits 8 GB may need Zyphra's fork", "ZAYA1-VL-8B, the vision model, has 35 times the downloads and isn't listed", "Zyphra's site lists a cloud product for its open models. Whether it serves this one wasn't checked", "Capability is pending because no independent results were looked up on the day", "unchecked on 10 November: the technical report (arXiv 2605.05365), the blog post, Zyphra's forks, its cloud product, Ollama's library and LM Studio's catalogue" ] }, "negative": 0, "verdict": "A small reasoning model trained from scratch by one lab, with a clean licence, 4-bit and FP8 builds and a team that answers questions. It's the hardest of the small models to run: its own architecture, a card that points at forks, and tool calling that needs a parser no release has. Little on limits, data or safety.", "bestFor": "People who want a small, fast reasoning model for maths and code and can live without tool calling", "strengths": [ "Apache 2.0, ungated", "5.9 GB at 4 bits, 760M parameters active", "Released Transformers and SGLang carry the architecture", "Zyphra replied in most of the recent discussions", "The original checkpoint is kept beside the reshaped one" ], "weaknesses": [ "Tool calling needs a parser that only Zyphra's vLLM fork has", "Released vLLM and llama.cpp don't run it, and there's no GGUF", "The card's install steps use forks", "No limitations, context length, languages, data or safety results on the card", "Benchmarks on the card are all from Zyphra's own harness" ], "agentNotes": [ "Use it for reasoning, maths and code. Don't count on tool calling outside Zyphra's vLLM fork", "Run it with Transformers 5.22.2 or SGLang 0.5.23, which have the model built in", "Use temperature 1.0 and top_p 0.95 in general, and temperature 0.6 for agent and code work", "Download `Zyphra/ZAYA1-8B`, not the legacy repository, for the released runtimes", "The quantised repositories have no licence tag. The main one is Apache 2.0" ], "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": 60, "reach": 80, "risk": 55, "upkeep": 50 }, "provenanceScore": 56, "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 Zyphra/ZAYA1-8B --reasoning-parser qwen3\n# the card's own route is vLLM from Zyphra's fork, with --tool-call-parser zaya_xml", "http": "# SGLang serves an OpenAI-compatible API at http://localhost:30000/v1. The card recommends temperature 1.0\n# and top_p 0.95 in general, and temperature 0.6 for agent and code work" }, "letme": { "capability": "https://letme.dev/model.open-weights", "tool": "https://letme.dev/zaya1-8b" }, "weights": { "trainedBy": "Zyphra", "builtOn": "ZAYA1-reasoning-base, Zyphra's own pre-trained model, per the card", "params": "8.4B", "active": "760M", "context": 131072, "goodFor": "Long step-by-step reasoning in maths and code on a small machine, and as the model inside a test-time compute harness.", "runtimes": [ "SGLang", "Transformers" ], "builds": [ { "name": "ZAYA1-8B-MXFP4-Experts", "url": "https://huggingface.co/Zyphra/ZAYA1-8B-MXFP4-Experts", "who": "publisher", "precision": "mxfp4", "sizeGB": 5.85, "memoryGB": 8, "format": "safetensors", "note": "Zyphra's build with the experts at 4 bits. Which released runtime loads it wasn't established" }, { "name": "ZAYA1-8B-FP8-Experts", "url": "https://huggingface.co/Zyphra/ZAYA1-8B-FP8-Experts", "who": "publisher", "precision": "fp8", "sizeGB": 9.64, "memoryGB": 12, "format": "safetensors", "note": "Zyphra's build with the experts in FP8" }, { "name": "ZAYA1-8B (BF16 safetensors)", "url": "https://huggingface.co/Zyphra/ZAYA1-8B", "who": "publisher", "precision": "bf16", "sizeGB": 17.68, "memoryGB": 24, "format": "safetensors", "runsOn": [ "SGLang", "Transformers" ], "note": "Four shards, reshaped for Transformers" } ], "run": "python -m sglang.launch_server --model-path Zyphra/ZAYA1-8B --reasoning-parser qwen3", "runBuild": "ZAYA1-8B (BF16 safetensors)", "checked": "2026-11-10", "minMemoryGB": 8 }, "family": { "lab": "Zyphra", "family": "ZAYA1", "url": "https://www.anchorterminal.com/open-weights/#zyphra", "others": [ { "name": "ZAYA1-VL-8B", "url": "https://huggingface.co/Zyphra/ZAYA1-VL-8B" }, { "name": "ZAYA1-74B-preview", "url": "https://huggingface.co/Zyphra/ZAYA1-74B-preview" } ] }, "notable": [ "9,474 downloads in the last 30 days and 598 likes on Hugging Face on 10 November 2026. Its vision sibling, ZAYA1-VL-8B, had 338,187 downloads (https://huggingface.co/api/models/Zyphra/ZAYA1-8B)", "Trained end to end by Zyphra, not fine-tuned from another lab's model, with 760M of its 8.4B parameters active per token, per the card (https://huggingface.co/Zyphra/ZAYA1-8B)", "The checkpoint was reshaped for a Transformers pull request, and the original is kept as ZAYA1-8B-legacy (https://huggingface.co/Zyphra/ZAYA1-8B)", "The card's install steps use Zyphra's forks of vLLM and Transformers. Released Transformers 5.22.2 and SGLang 0.5.23 now have the model; released vLLM doesn't (https://huggingface.co/Zyphra/ZAYA1-8B)", "The card says every number in its tables was run on Zyphra's own evaluation harness (https://huggingface.co/Zyphra/ZAYA1-8B)", "The FP8 and MXFP4 repositories carry no licence tag of their own (https://huggingface.co/api/models/Zyphra/ZAYA1-8B-MXFP4-Experts)", "A Zyphra member replied in 10 of the 16 most recent discussions (https://huggingface.co/api/models/Zyphra/ZAYA1-8B/discussions)" ], "area": "models", "details": [ { "label": "Surface graded", "value": "The open weights at huggingface.co/Zyphra/ZAYA1-8B and the FP8-Experts and MXFP4-Experts repositories beside it, read from the model card, the files and their metadata, the config, the chat template, the commit history and the runtimes' own code at their latest releases on 10 November 2026. Zyphra's home and about pages were read for who it is. The technical report, the blog post and Zyphra's forks weren't read. Nothing was downloaded or run" }, { "label": "Model", "value": "8,840,488,784 parameters in the files; the card says 8.4B with 760M active. A mixture of experts with 16 experts and one active per token, trained end to end by Zyphra and post-trained for reasoning from ZAYA1-reasoning-base. ZayaForCausalLM" }, { "label": "Files", "value": "Four safetensors shards in BF16 (17.68 GB), chat_template.jinja, config and tokenizer files. No pickle files and no Python. The FP8-Experts repository is 9.64 GB and the MXFP4-Experts one 5.85 GB. Hugging Face lists a SHA-256 for the large files" }, { "label": "Licence", "value": "Apache 2.0 in the main repository's metadata. There is no LICENSE file, and the FP8-Experts and MXFP4-Experts repositories have no licence tag" }, { "label": "Runtimes", "value": "The card documents vLLM and Transformers from Zyphra's own forks. Released Transformers 5.22.2 has the zaya model and SGLang 0.5.23 has srt/models/zaya.py. Released vLLM 0.33.0 and llama.cpp b12242 have no code for it" }, { "label": "Context", "value": "Not stated on the card. config.json gives 131,072 positions" }, { "label": "Tools and reasoning", "value": "The card's vLLM command sets auto tool choice with a zaya_xml tool-call parser, which exists in Zyphra's fork and in no released runtime, and the qwen3 reasoning parser. The chat template takes tools and an enable_thinking switch" }, { "label": "Hardware", "value": "No memory figure on the card, which says the model can run on-device. Its parallel-deployment note covers 8 GPUs. The 4-bit build is 5.85 GB" }, { "label": "Training data", "value": "Not described on the card. The technical report wasn't read" } ], "provenance": { "legalEntity": "Zyphra, named on the card and its site. The legal entity wasn't looked up", "domain": "zyphra.com", "domainRegistered": "", "endpointOnVendorDomain": null, "terms": "", "privacy": "", "statusPage": "", "changelog": "https://huggingface.co/Zyphra/ZAYA1-8B/commits/main", "securityTxt": "none", "checked": "2026-11-10", "notes": [ "www.zyphra.com's robots.txt allows reading outside /studio and /api. Its about page gives a San Francisco address and names no people in the text read.", "The domain's registration date wasn't looked up. www.zyphra.com has no security.txt (404 on 10 November 2026).", "Nothing is hosted under this listing, so Apache 2.0 stands in for terms of service and no privacy policy is scored. Zyphra's site lists a cloud product for its open models, which wasn't read.", "The changelog is the commit history, 41 commits from 4 May to 26 June 2026." ], "score": 56, "checks": [ { "check": "Legal entity named", "value": "Zyphra, named on the card and its site. The legal entity wasn't looked up", "points": 20, "max": 20, "state": "ok" }, { "check": "Domain age", "value": "zyphra.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 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": "not researched yet", "points": 0, "max": 0, "state": "na" }, { "check": "Support route", "value": "help pages, a community", "points": 5, "max": 5, "state": "ok" } ] }, "humans": { "slug": "zaya1-8b", "checked": "2026-11-10", "company": "Zyphra", "peopleNote": "Zyphra's about page, read on 10 November 2026, gives an address in San Francisco and named no people in the text we could read. The model card names none either. Ownership wasn't looked up.", "companySocial": { "github": "https://github.com/Zyphra" }, "support": { "urls": [ "https://www.zyphra.com" ], "community": [ { "label": "Hugging Face discussions", "url": "https://huggingface.co/Zyphra/ZAYA1-8B/discussions", "kind": "forum" } ] }, "sources": [ "https://huggingface.co/Zyphra/ZAYA1-8B", "https://www.zyphra.com/about", "https://www.zyphra.com/" ], "ownership": { "status": "unknown", "checked": "2026-11-10" } }, "pageJsonUrl": "https://www.anchorterminal.com/tools/zaya1-8b.json" } }