Qwen3.6-35B-A3B

# Fix list: Qwen3.6-35B-A3B

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/qwen3-6-35b-a3b, the October 2026 research run, assessed 21 October 2026. Grade B, 63.2 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 Qwen3.6-35B-A3B: 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, 58 out of 100, up to 14.7 more on the total

Why it scored 58: From 14 independent benchmark results in 5 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, Long context 81, Instruction following 78, Reasoning and knowledge 57, Coding 55, Tool calling and agent tasks 46. 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: File safety earns 25 for safetensors only, loaded by the runtimes' own code and a Jinja template with nothing run from the repository. Provenance earns 15, files in the Qwen organisation the publisher links and a hash for every file. Safety evaluation and agent guidance earn 0, since the card has neither safety results nor any guidance on refusals, injection or safeguards. Licence risk keeps all 15, as Apache 2.0 carries none of the four conditions and the late LICENSE file only wrote down what the card already said. The security programme earns 0, with no reporting route found on qwen.ai, the card or the GitHub repository, which is noted as unchecked for Alibaba's group-wide centre.

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, 36 out of 100, up to 6.4 more on the total

Why it scored 36: The model card earns 11, covering intended use, architecture, parameters and context but not limitations, languages or a cut-off. Training data earns 5 for a general description with no cut-off. The last change earns 10, a README edit on 24 April 2026, 180 days ago, so it drops to 5 on the next day without a new commit. Responsiveness earns 0, with no Qwen reply in 61 discussions. Version policy earns 5, the earlier version kept in its own repository without a note. File health earns 5, the files agree but reports about tool calling and the template wait unanswered.

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. Reach, 85 out of 100, up to 3.8 more on the total

Why it scored 85: Memory earns 20 because the smallest 4-bit build from a runtime's own library is Ollama's Q4_K_M at 23 GB, which comes to 27.6 GB with the fifth added, so it needs a 32 GB machine rather than a 24 GB one. Runtimes earn the full 25, since released llama.cpp, Ollama, vLLM, SGLang and Transformers all run the architecture. Builds earn 10 for Alibaba's one quantised build, FP8, with no 4-bit or GGUF from the publisher. Terms earn 15 for plain Apache 2.0, download earns 10 for ungated files, and hosted earns 5 because DeepInfra, Featherless AI and Scaleway serve it through Hugging Face.

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.

## 5. Agent interface, 70 out of 100, up to 3 more on the total

Why it scored 70: The chat template earns 20 as a Jinja file in the repository. The tool-calling format earns 10, because the card names the qwen3_coder parser but leaves the return format to the template. Trained tool calling earns 20, with several calls to a turn in the template. Structured output earns 0, as JSON output isn't documented for the model. Reasoning control earns 10, thinking off per request with enable_thinking and the reasoning returned in think tags for the qwen3 parser. Context earns 10 for 262,144 tokens native.

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 and scored as absent, whether Alibaba's group-wide security response centre takes reports about Qwen models; no route from the card, qwen.ai or QwenLM on GitHub was found
- Unchecked, issues on QwenLM/Qwen3.6, since the GitHub API wasn't reachable; responsiveness is read from Hugging Face discussions only
- Unchecked, the qwen.ai release blog, a JavaScript app with no readable text, which may hold safety, data or language details the card lacks
- Last change sits at the 180-day edge and drops from 10 to 5 on 22 October 2026 without a new commit
- Alibaba Cloud Model Studio serves it per the card but has no listing, so no sameModel is set

## Weaknesses

- No safety evaluation, no agent-use guidance and no way to report a security flaw
- No reply from a Qwen member in any of its 61 discussions, including reports on tool calling and the template
- No 4-bit or GGUF build from Alibaba, only FP8
- Card leaves out limitations, languages, knowledge cut-off and training data
- Structured output isn't documented, and the files haven't changed since 24 April 2026

## 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 `ollama run qwen3.6:35b` (Q4_K_M, 23 to 24 GB on disk, about 28 GB in memory)
- On the reference machine use `vllm serve Qwen/Qwen3.6-35B-A3B --max-model-len 131072 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder`, BF16, since vLLM doesn't document FP8 compute on gfx1151
- Use the qwen3_coder tool-call parser and the qwen3 reasoning parser in vLLM 0.19.0 or SGLang 0.5.10 or later
- Thinking is on by default and /nothink isn't supported, so pass enable_thinking as false in chat_template_kwargs for short tasks, or preserve_thinking as true to keep earlier reasoning
- Card sampling for thinking is temperature 1.0, top_p 0.95, top_k 20, presence_penalty 1.5, and for non-thinking temperature 0.7, top_p 0.8
- Keep at least 128K of context as the card advises, and turn on YaRN only beyond 262,144 tokens
- Validate JSON output yourself or use the runtime's guided decoding, since the model card documents none

## What the review panel asked for

- successor note on Qwen3.5
- a publisher GGUF build
- a tool-calling section on the card
- publish safety results
- 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.