Qwen3.8-Flash-Next

# Fix list: Qwen3.8-Flash-Next

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/qwen3-8-flash-next, the October 2026 research run, assessed 21 October 2026. Grade C, 61.8 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 Qwen3.8-Flash-Next: 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, 45 out of 100, up to 11 more on the total

Why it scored 45: File safety 25, safetensors only and nothing from the repository runs to load it or build prompts. Provenance 15, the files are in the Qwen organisation that Qwen's GitHub links to, and every file has a checksum. Safety evaluation 0, none published on the card or in the technical report. Agent guidance 0, nothing on prompt injection, refusals or safeguards. Licence risk 5 of 15, 5 off for the field-of-use limit on model-as-a-service and coding or office assistant businesses and 5 off for the naming duty above a user or revenue threshold. Security programme 0, no security.txt, disclosure policy or security contact for Qwen was found, and Alibaba's group security centre couldn't be checked, so it's scored as absent.

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. Capability, 74 out of 100, up to 9.1 more on the total

Why it scored 74: From 7 independent benchmark results in 4 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 90, Coding 76, Reasoning and knowledge 75, Tool calling and agent tasks 67. 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.

## 3. Reach, 65 out of 100, up to 8.8 more on the total

Why it scored 65: Memory 5 of 30, the smallest listed 4-bit build is Ollama's 120 GB q4_K_M, 144 GB with a fifth added, which is beyond the reference machine and inside 256 GB. Ollama's Apple-only 105 GB MLX build isn't among the builds and would land in the same band. Runtimes 25, released versions of llama.cpp, Ollama, vLLM, SGLang and Transformers all run the qwen4_exp architecture. Builds 10 of 15, Alibaba's only quantised build is FP8. Terms 10 of 15, a custom licence that's free for commercial use except for model-as-a-service and coding or office assistant businesses, which need a separate licence. Download 10, neither repository is gated. Hosted 5, Alibaba's Qwen Cloud serves it as Qwen3.8-Flash, which the card says is based on these weights and OpenRouter maps to this repository.

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.

## 4. Documentation and upkeep, 36 out of 100, up to 6.4 more on the total

Why it scored 36: Model card 11 of 20, intended use, architecture, parameters and context are there, but limitations, languages and a knowledge cut-off aren't. Training data 0, not described beyond a token count relative to an earlier model. Last change 15, the FP8 config fix on 31 August 2026, 51 days ago. Responsiveness 0, Qwen answered none of the 40 most recent questions and reports, its only reply came in the first week. Version policy 5, a first release with nothing replaced. File health 5, the files agree with the card, but a 19 October report about the template waits without a reply.

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.

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

Why it scored 70: Chat template 20, a Jinja file in the repository. Tool-calling format 10 of 25, the format lives only in the template and the parser name only in serving commands, and Alibaba doesn't describe how tools are declared or returned. Trained tool calling 20, the template renders several calls in one turn. Structured output 0, not documented for the weights. Reasoning control 10, thinking turns off per request, reasoning_effort budgets it and the card shows the reasoning coming back in reasoning_content. Context 10, 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.

- Hosted too rests on Qwen Cloud's Qwen3.8-Flash, which the card calls the official version based on this model with 1M context by default and built-in tools. It isn't stated to be the same weights, and Hugging Face lists only one provider (Featherless AI) for the repository, so if the founder reads Qwen3.8-Flash as a different model the line drops to 0
- unchecked, issues on github.com/QwenLM/Qwen3.8-Flash-Next, since GitHub's API isn't enabled for this repository in this session, so responsiveness rests on the Hugging Face discussions only
- unchecked, whether Alibaba's group security response centre takes reports about Qwen models, no link from the card, README or qwen.ai was found, so the security programme is scored as absent
- hardware.json gives Ollama's 105 GB 125b-mlx tag no precision, and the tags page shows it is the same digest as 125b-a6b-nvfp4, a 4-bit build for Apple machines only
- Capability isn't scored here, and confidence stays medium until the script's independent results are in

## Weaknesses

- No build fits the reference machine's 107.4 GB, and Ollama's smallest Linux build is 120 GB
- The Qwen Community License requires a separate licence before commercial use by a model-as-a-service or coding or office assistant business
- No safety results, agent guidance, training data, limitations, languages or knowledge cut-off published
- Labelled an experimental preview, and Qwen has answered none of the 40 most recent discussions
- The tool-calling format is only in the template, and structured output isn't documented

## 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.

- On a 128 GB Mac start it with `ollama run qwen3.8-flash-next:125b-mlx` (105 GB, NVFP4), on Linux the smallest Ollama build is `qwen3.8-flash-next:125b-a6b-q4_K_M` at 120 GB, which won't fit a 107.4 GB GPU share
- On a server, `vllm serve Qwen/Qwen3.8-Flash-Next-FP8 --tensor-parallel-size 4 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder`
- Use the qwen3_coder tool parser and the qwen3 reasoning parser, as Qwen's GitHub README does
- Thinking is on by default, pass `chat_template_kwargs` with `enable_thinking` false to skip it, and read the reasoning from `reasoning_content`, even though SGLang's cookbook says thinking can't be turned off
- `reasoning_effort` takes xhigh (default), medium or low, and any other value makes the template raise an error
- Send at most one system message and put it first, since the template raises an error otherwise
- Sampling per the card is temperature 1.0 and top_p 0.95 when thinking, temperature 0.7, top_p 0.8 and presence_penalty 1.5 without
- vLLM keeps the 51B n-gram table in host memory on 80 GB GPUs by default, per its recipe
- Validate JSON output yourself or use the runtime's guided decoding, since structured output isn't documented
- Read the Qwen Community License before serving it to others as an API or inside a coding or office assistant product

## What the review panel asked for

- answer the template report
- state a version policy
- a price for the separate licence
- confirmation that Qwen3.8-Flash is these weights
- a tool-calling section
- align the SGLang cookbook
- a disclosure policy
- safety 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.