InternVL3.5-GPT-OSS-20B-A4B-Preview

# Fix list: InternVL3.5-GPT-OSS-20B-A4B-Preview

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/internvl3-5-gpt-oss-20b-a4b-preview, the October 2026 research run, assessed 21 October 2026. Grade D, 51.2 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 InternVL3.5-GPT-OSS-20B-A4B-Preview: 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. Reach, 60 out of 100, up to 15.4 more on the total

Why it scored 60: Memory 15 of 30, the only builds are BF16 at 42.5 GB, or 51 GB with a fifth added, which fits 64 GB. Runtimes 20, Transformers and vLLM load this repository, and SGLang and llama.cpp run the same weights in their GitHub format. Builds 0, OpenGVLab publishes no quantised build. Terms 15, Apache 2.0 with nothing added by the bases. Download 10, not gated. Hosted 0, no provider serves it.

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.

## 2. Risk, 55 out of 100, up to 13.8 more on the total

Why it scored 55: File safety 25, safetensors only and no code in the repository. Provenance 15, the files are in the OpenGVLab organisation its GitHub links, with a checksum on every file. Safety evaluation 0 and agent guidance 0, nothing published. Licence risk 15, Apache 2.0 over Apache and MIT bases, with gpt-oss's usage policy asking only for lawful use. Security programme 0, no policy, security.txt or advisories.

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, 28 out of 100, up to 11.1 more on the total

Why it scored 28: Model card 8 of 20, the family card misses limitations, context, languages and a cut-off. Training data 5, described without a cut-off. Last change 0, nothing has changed in over a year. Responsiveness 10, OpenGVLab answered 3 of 8 questions across the two repositories. Version policy 5, a first release. File health 0, the template uses markers the tokenizer doesn't have and contradicts the twin's template, and the card names the wrong language model in its licence line.

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, 45 out of 100, up to 8.5 more on the total

Why it scored 45: Chat template 20, a Jinja file in the repository, though it doesn't match the tokenizer. Tool-calling format 10 of 25, only the twin's inherited harmony template mentions tools. Trained tool calling 0, no evidence it was trained to call tools. Structured output 0. Reasoning control 10, a system prompt turns thinking on and the card says the reasoning comes back in think tags. Context 5, trained at 32K tokens.

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.

- SGLang and llama.cpp are credited for running the same weights in OpenGVLab's GitHub format, not this repository's Transformers format; on a strict reading of this repository alone, runtimes is 10
- Whether the model was trained on the ChatML-style markers in this repository's template or on gpt-oss's harmony format isn't documented, so which template is right is unknown
- 620,226 downloads in 30 days against 10 likes; the source of the downloads isn't known

## Weaknesses

- No quantised build from OpenGVLab, so about 51 GB at BF16
- Chat template uses markers its tokenizer lacks and differs from the twin's harmony template
- No change to the files since September 2025, and it's still a preview
- No safety results, agent guidance or security contact
- No documented tool calling or structured output, and trained at 32K tokens

## 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.

- Serve with `vllm serve OpenGVLab/InternVL3_5-GPT-OSS-20B-A4B-Preview-HF --max-model-len 32768`, which is our command; the card gives none for vLLM
- Check the rendered prompt before relying on it, since the shipped template's <|im_start|> markers aren't tokens in the gpt-oss tokenizer; the GitHub-format repository's harmony template is the alternative
- Use Transformers 4.55.0 or later, as the card requires for this size
- For thinking, send the card's thinking system prompt and parse the reasoning out of the <think> tags, with temperature 0.6 and sampling on as the card advises
- Keep inputs within 32K tokens, the length it was trained at, whatever the config says

## What the review panel asked for

- fix or retire the preview
- an MXFP4 build of the experts
- state which prompt format it was trained on
- document the trained prompt format
- 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.