Qwen3-VL-4B-Instruct

# Fix list: Qwen3-VL-4B-Instruct

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/qwen3-vl-4b-instruct, the October 2026 research run, assessed 21 October 2026. Grade D, 52.6 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-VL-4B-Instruct: 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. Capability, 25 out of 100, up to 26.3 more on the total

Why it scored 25: From 10 independent benchmark results in 6 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, Images 44, Instruction following 38, Coding 32, Reasoning and knowledge 18, Long context 16, Tool calling and agent tasks 12. 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.

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

Why it scored 55: Safetensors and GGUF only and no code in the repositories, so the FP8 card's trust_remote_code flag runs nothing (25). Files in the Qwen organisation with a checksum on every file (15). No safety evaluation at all (0). No guidance for agent use, which matters for a model sold as a GUI agent (0). Apache 2.0 carries none of the licence risks (15). No security contact, disclosure policy or public advisories found for Qwen (0), and Alibaba's group-wide 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.

## 3. Documentation and upkeep, 31 out of 100, up to 6.9 more on the total

Why it scored 31: The card misses limitations, languages and a cut-off (11). Training data is described only in the technical report, with no cut-off (5). The last change to code that runs it was the qwen-vl-utils fix of 4 January 2026, and the weights haven't changed since November 2025 (5). Qwen answered none of the 14 Hugging Face discussions that aren't pull requests (0), and GitHub issues weren't counted. Earlier versions kept in their own repositories without notes (5). Two 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.

## 4. Agent interface, 60 out of 100, up to 4 more on the total

Why it scored 60: A Jinja chat template in the repository (20). The tool-call format lives only in the template, Hermes-style JSON in tool_call tags, and Alibaba names no parser (10). Trained for function calling with several calls allowed in a turn, per the family's technical report (20). No documented structured output (0). No thinking mode in this edition, so reasoning control scores 0 by the rule. 262,144 tokens of native context (10).

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.

## 5. Reach, 95 out of 100, up to 1.3 more on the total

Why it scored 95: Alibaba's own Q4_K_M GGUF with its vision file is 2.95 GB, so with a fifth added it needs about 3.5 GB and fits well within 16 GB (30). Transformers 4.57.0, vLLM, SGLang, llama.cpp and Ollama all run it in released versions, which reaches the 25 cap. Alibaba publishes FP8 and GGUF builds at F16, Q8_0 and Q4_K_M (15). Apache 2.0 with nothing added (15). No gate and no account (10). Only Featherless was found serving it, so it isn't hosted too (0).

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

Whether an agent can run it, and where.

- 0 to 30, memory. The smallest build from the publisher or a runtime's own library that keeps 4-bit precision or better, at its file size plus a fifth for the runtime and a working context. 30 for 16 GB or less, 25 for 24 GB, 20 for 32 GB, 15 for 64 GB, 10 for what fits our reference machine (the GPU share of 128 GB of unified memory), 5 for 256 GB, 0 above.
- 0 to 25, runtimes. 5 for each of llama.cpp, Ollama, vLLM, SGLang, LM Studio and Hugging Face Transformers that runs the architecture in a released version, from the runtime's own code, library or documentation. A publisher's fork or an unreleased branch doesn't count. 25 at most.
- 0 to 15, builds from the publisher. 15 for quantised builds at two precisions or more (FP8, 4-bit, GGUF), 10 for one, 0 when only other people's quantisations exist.
- 0 to 15, terms of use. 15 for an OSI-approved licence with nothing added, 10 for a custom licence that lets anyone use it commercially for free below a threshold or under a use policy, 5 for research or non-commercial use only, 0 for no licence or unclear terms.
- 0 to 10, download. 10 when the files download with no account and no terms to accept, 5 for a click-through that needs an account, 0 when a person approves each request.
- 5, hosted too. The publisher's own API serves it, or at least two inference providers do, so an agent without the hardware can still call it.

When our reference-machine runs start, speed on that machine (tokens per second and time to first token at the build that fits) joins Reach.

## 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, responsiveness on QwenLM/Qwen3-VL GitHub issues. Only Hugging Face discussions were counted. A count of GitHub issues could raise the line
- unchecked, whether Alibaba's group-wide security response centre takes reports about Qwen models. Nothing on the card, README or qwen.ai links to one, so the security programme scores 0
- unchecked, whether Alibaba Cloud Model Studio serves this exact model today. Its model list renders with script and couldn't be read. If it does, hosted too is met (5)
- Reasoning control scores 0 because this is an Instruct edition with no thinking, and the checklist has no not-applicable rule

## Weaknesses

- No safety evaluation, agent-use guidance or security contact
- Tool calling isn't documented beyond the template, and Alibaba names no parser
- No thinking mode in this edition and no documented JSON output
- No reply from Qwen to any of the 14 Hugging Face discussions that aren't pull requests
- The card leaves out limitations, languages, knowledge cut-off and memory needs
- Only Featherless was found serving it for agents without the hardware

## 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.

- Start it with `llama-server -hf Qwen/Qwen3-VL-4B-Instruct-GGUF:Q4_K_M --jinja -c 32768`, or `ollama run qwen3-vl:4b-instruct`
- In Ollama the plain `qwen3-vl:4b` tag is the Thinking edition, so name `4b-instruct`
- With the GGUF, load the mmproj vision file too, or images won't be read
- Parse tool calls as JSON inside `<tool_call>` tags, or use SGLang's `--tool-call-parser qwen`
- Use the card's sampling, `temperature=0.7, top_p=0.8, top_k=20, presence_penalty=1.5` for vision and `temperature=1.0, top_p=1.0, top_k=40, presence_penalty=2.0` for text
- Validate JSON output yourself, since structured output isn't documented
- vLLM needs 0.11.0 or later and Transformers 4.57.0 or later

## What the review panel asked for

- answer the header and state_dict reports
- memory needs on the card
- document the tool format
- guidance on on-screen instructions
- a security contact

## 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.