Qwen3-Embedding-8B

# Fix list: Qwen3-Embedding-8B

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/qwen3-embedding-8b, the October 2026 research run, assessed 21 October 2026. Grade BB, 74.7 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Qwen3-Embedding-8B: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Risk, 55 out of 100, up to 9 more on the total

Why it scored 55: Safetensors and GGUF only, loaded with no code from the repository (25). Files in the Qwen organisation with a checksum on every file (15). No safety evaluation (0). No guidance for agents on untrusted documents or retrieved text (0). Apache 2.0 carries none of the licence risks, and the LICENSE file added in July 2025 matches what the metadata said from the start (15). No security contact, disclosure policy or advisories for Qwen (0), and Alibaba's group security centre wasn't checked.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-risk):

What could go wrong when you run it, and what the publisher has done about it. A higher score means less risk.

- 0 to 25, file safety. 25 for weights in safetensors or GGUF only, 15 when pickle files (.bin, .pt, .pth) sit beside them, 0 for pickle only. 10 off when the documented way to load it or to build its prompts runs code from the repository (`trust_remote_code`, or the publisher's scripts in place of a template).
- 0 to 15, provenance. 10 when the files come from the publisher's own organisation, the one its site or docs link to, and 5 for a checksum on every file.
- 0 to 20, safety evaluation. 20 when results are published and someone other than the publisher ran them or can rerun them, 10 for the publisher's own results, 0 for none.
- 0 to 15, guidance for agent use. 15 for documented prompt-injection guidance or an instruction hierarchy the model was trained to follow, 10 for a statement of what its trained refusals cover and the safeguards to add around it, 5 for general cautions, 0 for none.
- 0 to 15, licence risk. 15 to start, 5 off for each of these, a use policy or field-of-use limits beyond the law, terms the licensor can change or revoke, a revenue or user threshold, and a licence that changed without a dated note or whose link is broken. Not below 0.
- 0 to 10, a security programme. 5 for a way to report a flaw (security.txt, a disclosure policy or a security contact), 5 for advisories handled in public.

When our reference-machine runs start, measured robustness (prompt injection with AgentDojo, over-refusal with XSTest) joins Risk.

## 2. Documentation and upkeep, 21 out of 100, up to 7.9 more on the total

Why it scored 21: The card misses limitations, named languages and a cut-off (11). Training data is described only in the technical report, with no cut-off (5). Nothing that runs the model has changed since July 2025 (0). Qwen answered none of the 19 questions and reports on Hugging Face (0), and GitHub issues weren't counted. GGUF files and the tokenizer were replaced in place without a note (0). The architecture name in config.json doesn't match the shards, and NaN and dimension reports wait without a reply (5).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-upkeep):

- 0 to 20, the model card. Intended use, limitations, architecture, parameters, context length, languages and knowledge cut-off. 20 for all seven, 3 off for each that's missing.
- 0 to 20, training data. 20 for the data published or listed with sources and sizes, 15 for a data card that names the sources or the mix, the cut-off and the filtering, 10 for a general description with a cut-off, 5 without one, 0 for none.
- 0 to 20, the last change to the files, the template or the reference code. 20 within 30 days, 15 within 90, 10 within 180, 5 within a year, 0 older.
- 0 to 20, responsiveness. The publisher's replies to the 40 most recent questions and reports in the repository's discussions or issues. 20 for most, 10 for some, 5 for a few, 0 for none.
- 0 to 10, version policy. 10 for earlier versions kept in their own repositories with a dated note when one is replaced, 5 for earlier versions kept without notes or a first release, 0 when files were replaced in place without a note.
- 0 to 10, file health. 10 when config, tokenizer, template and card agree and no report of a broken file waits without a reply, 5 when one does, 0 when the files contradict the card.

## 3. Capability, 86 out of 100, up to 4.9 more on the total

Why it scored 86: From 4 independent benchmark results in 2 areas, each scored as its share of the way from the benchmark's floor to the best result any model has on the same leaderboard. By area, Other MTEB task types 87, Retrieval 85. Publisher-submitted results count at half weight, and numbers that appear only on the publisher's card don't count.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-capability):

How well the model does on the benchmarks that matter to an agent. We don't average a publisher's own numbers into a grade. A result counts in full when we ran it on our reference machine, or when an independent party ran it and published the method (the Berkeley Function Calling Leaderboard, Artificial Analysis, Epoch AI's Benchmarking Hub, the SWE-bench and Terminal-Bench leaderboards, the Open ASR Leaderboard and others that run models themselves). A result the publisher submitted to such a leaderboard counts at half weight. A number that appears only on the publisher's card is shown on the listing as the publisher's claim and doesn't count.

Each result is scored as its share of the way from the benchmark's floor (chance for multiple choice, zero for a pass rate) to the best result any model, open or closed, has on the same leaderboard and setting on the day we read it. A model that matches the best model anywhere scores 100 on that benchmark. For an error rate such as word error rate, lower is better and the scale runs the other way, and where the leaderboard gives no floor the rate is read against the best model's, so twice the best model's error rate scores 50. Speed as a multiple of real time (RTFx) runs from about 1 to over 20,000, so it's read on a log scale from real time to the fastest model. On the MTEB leaderboard, where most models score between 0.5 and 0.75, the floor is the median of every model the same tab lists, and its results count at half weight because MTEB doesn't say who ran them and most are submitted with the model. An arena rating (Elo) isn't a ratio, so its floor is 400 points below the best model on that arena, the gap at which the better model wins about ten games in eleven. Where a model has results at several settings (reasoning effort, say), its best one counts, because that's the setting an agent would run.

The results group into areas, each weighted within the category.

- Tool calling and agent tasks, 35 (BFCL, τ²-bench, Terminal-Bench and agent suites like them)
- Coding, 20 (SWE-bench Verified, LiveCodeBench, Aider Polyglot)
- Reasoning and knowledge, 20 (GPQA Diamond, MMLU-Pro, Humanity's Last Exam, composite indexes that publish their components)
- Instruction following, 15 (IFEval, IFBench)
- Long context, 10 (RULER, LongBench and long-context retrieval tests)

A speech model is read on its own areas, recognition accuracy (word error rate on the Open ASR Leaderboard's sets, 60) and speed (real-time factor, 40) for speech to text, and intelligibility and speaker similarity (Seed-TTS-eval, the Open TTS Leaderboard, 70) and speed (30) for text to speech. A publisher's own converted copy of the same weights (Qwen's -hf builds, say) counts as the same checkpoint. An embedding model is read on retrieval (60) and the other MTEB task types, classification, clustering, reranking and similarity among them (40), from the MTEB leaderboard. A vision model adds an image area (MMMU, ChartQA, DocVQA or OCRBench) at 20, and the others shrink to fit.

Capability is the weighted mean of the areas with results, and needs results in at least two areas (one for a speech or embedding model). With less than that it's pending for that listing, its weight is shared across the other four, and the total is capped at 61, so a model nobody has benchmarked independently can't reach B, and every model with results ranked B or better stays above it. The listing says so. Until 21 October 2026 the cap was 69, which put models nobody had benchmarked above ones that had been.

## 4. Agent interface, 65 out of 100, up to 3.5 more on the total

Why it scored 65: Read for an embedding model, the tool-calling and structured-output lines stand for what it takes and returns. The card documents the instruction format, pooling, normalisation, padding and dimensions, with code for four runtimes and the llama.cpp flags (25). It was trained to take one task instruction per query, one text to one vector (10). Qwen's repository shows how to ask for a shorter vector in Transformers and vLLM (15). The chat template scores 10 because the shipped Qwen3 template isn't the prompt path, which lives in the card's code and the Sentence Transformers prompt. No reasoning to control (0) and 32K trained context (5).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-interface):

What a model and a developer need to talk to it. Speech, image and other models with no chat interface read the tool-calling and structured-output lines for what the model takes and returns (voices, timestamps, streaming, typed outputs), and the note says how.

- 0 to 20, the chat template. 20 for a template in the repository (Jinja, in the tokenizer config or its own file), 10 when the prompt format is only given as code you run.
- 0 to 25, the tool-calling format. 25 when the publisher documents how tools are declared and how calls come back, with the parser to use in a named runtime, 10 when only the template or a runtime's own recipe says it.
- 0 to 20, tool calling the model was trained for. 20 with several calls in one turn, 10 for one call at a time.
- 0 to 15, structured output. 15 when JSON or JSON Schema output is documented for the model with how to ask for it, 5 when the card claims it without saying how.
- 10, reasoning control. Thinking that can be turned off or budgeted per request, with where the reasoning text comes back.
- 0 to 10, context. 10 for 128K tokens or more trained, 5 for 32K to 127K. Context reached only by scaling the position encoding counts one band down.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- Unchecked, GitHub issues on QwenLM/Qwen3-Embedding, since the GitHub API isn't enabled for this session; responsiveness rests on Hugging Face discussions
- Unchecked and scored as absent, whether Alibaba's group security response centre takes reports about Qwen models
- The chat template line reads the shipped Qwen3 template as not the prompt format, since every documented path builds the Instruct and Query string in code; a literal reading of the line would give 20
- Trained tool calling reads one instruction and one text per vector as the one-at-a-time band; the line has no clear embedding equivalent of several calls in a turn
- Training data reads the report's Table 6 as a description without a cut-off (5); a reader who takes its named sets and stage sizes as a listing with sources and sizes would give 20
- No Qwen or Alibaba Cloud API was found that says it serves these weights, and none is listed here, so no sameModel is set

## Weaknesses

- No change to the files or reference code since July 2025, and no reply from Qwen to any of its 19 questions
- GGUF files and the tokenizer were replaced in place without a note
- No safety evaluation, agent guidance or security contact
- 32K tokens of context, and the card's Transformers example truncates at 8,192
- Open reports of NaN embeddings in float16 and a config that names a head the shards don't hold

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Quickest start is `ollama pull qwen3-embedding:8b`, then POST to /api/embed
- With llama.cpp use Qwen's GGUF and `llama-server -hf Qwen/Qwen3-Embedding-8B-GGUF:Q4_K_M --embedding --pooling last -ub 8192`, the flags the GGUF card gives
- Prefix each query with `Instruct: <one-sentence task>\nQuery:` and embed documents with no prefix, the card says leaving it off costs 1 to 5 per cent
- Write instructions in English even for other languages, as the card advises
- To shorten vectors, keep the first N values and L2-normalise again; in vLLM a dimensions request needs `--hf-overrides '{"is_matryoshka": true}'` because config.json doesn't declare it
- Pad on the left if you pool yourself, and run in BF16 rather than float16, where a discussion reports NaN outputs
- Treat retrieved documents as untrusted text before passing them to a model, since nothing in the card covers poisoning or injection

## What the review panel asked for

- correct config.json
- memory figures on the card
- answer the float16 reports
- retrieval-poisoning guidance
- publish bias results

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.