Ming-3.0-audio-tiny

# Fix list: Ming-3.0-audio-tiny

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/ming-3-0-audio-tiny, the October 2026 research run, assessed 21 October 2026. Grade D, 49.8 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 Ming-3.0-audio-tiny: 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, 45 out of 100, up to 21.2 more on the total

Why it scored 45: Memory 20, the only build inclusionAI publishes is BF16 at 20.78 GB, which with a fifth on top is 24.9 GB, just past the 24 GB band, so it lands in the 32 GB band. Runtimes 0, none of the six named runtimes runs the audio architecture in its own released code, since vLLM and SGLang need inclusionAI's branches, Transformers needs the repository's remote code, llama.cpp knows only the text backbone, and Ollama and LM Studio have no entry. SGLang-Omni serves it from main but isn't on the list. Builds 0, no quantised build from inclusionAI. Terms 15, plain MIT, with a bundled fastText file under its own CC BY-SA 3.0 notice. Download 10, no gate. Hosted 0, no inference provider on Hugging Face and no inclusionAI API found.

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, 50 out of 100, up to 15.4 more on the total

Why it scored 50: File safety 15, safetensors only with no pickle weights, less 10 because every documented way to load it runs the repository's code, and 40 compiled .pyc files sit under talker/. Provenance 15, the files are in inclusionAI's own organisation, which its GitHub repository and card link, and every file has a checksum. Safety evaluation 0, none published. Agent guidance 0, the only caution is about reviewing remote code before loading it. Licence risk 15, MIT with no policy, threshold or revocable terms, and the card's licence link resolves. Security programme 5, for Ant Group's AntSRC reporting centre, read as a company-wide route the way Google's was for Gemma, with no SECURITY.md, security.txt or advisories found.

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, 46 out of 100, up to 8.3 more on the total

Why it scored 46: Model card 11, it covers intended use, architecture, parameters and languages and leaves out limitations, context length and a knowledge cut-off. Training data 0, not described anywhere. Last change 20, the files changed on 19 October and the reference serving code on 20 October 2026. Responsiveness 0, the only discussion is a merged pull request and nobody outside inclusionAI has asked anything yet, two days after release, and GitHub issues couldn't be checked. Version policy 5, a first release. File health 10, config, template, card and profiles agree and no broken-file report waits.

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, 65 out of 100, up to 5.4 more on the total

Why it scored 65: Read for open weights and, as a speech model, for what it takes and returns, while this model also has a chat interface with function calls. Chat template 20, a Jinja file in the repository. Tool-calling format 10, the format of tool calls comes only from the template and no parser is named, although inclusionAI does document its audio inputs, voices, cloning and output modalities for its own serving branches. Trained tool calling 20, the template renders several calls in one turn and inclusionAI reports function-calling results for Tiny. Structured output 5, transcripts come back in a documented typed form (language label, tab, text) and speech as WAV or streamed PCM, but no JSON, JSON Schema or timestamps are documented. Reasoning control 0, thinking can be turned on per request but nothing says where the reasoning text comes back. Context 10, 131,072 positions in config.json with no scaling, matched by the publisher's vLLM-Omni profiles.

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.

- The tool-calling format is read for the model's text tool calls (10), as for Ming-3.0-audio-flash, since the checklist reads a speech model for what it takes and returns only when it has no chat interface. Read purely as a speech model, both audio listings would earn 25
- Couldn't check issues on github.com/inclusionAI/Ming-3.0-audio-flash, since the GitHub API and pages weren't open to this session. Responsiveness scores 0, partly because nobody has asked on Hugging Face yet two days after release
- Memory is read on the whole BF16 repository (20.78 GB). The thinker shards alone (17.7 GB, 21.3 GB with headroom) would land in the 24 GB band for text and ASR without the Talker
- Whether SGLang-Omni's ming3_audio pipeline is in a tagged release; it is on main, and SGLang-Omni isn't a named runtime either way
- Capability is pending until independent results exist; only the publisher's own ASR and audio tables were found
- Whether AntSRC's scope covers inclusionAI's open models wasn't checked, since the site is a script-rendered app. The 5 rests on it being Ant Group's company-wide reporting route, read the same way for all five Ming-3.0 listings

## Weaknesses

- No named runtime runs it in a released version, only SGLang-Omni main and inclusionAI's own branches
- Every loading path runs repository code, and 40 compiled .pyc files ship under talker/
- BF16 only from inclusionAI, about 25 GB with headroom, and no inference provider
- No tool-call parser named, no JSON or timestamps documented, and no word on where reasoning text comes back
- No safety evaluation, agent guidance, training-data description, limitations or cut-off, and no security route of its own

## 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.

- Fastest start is SGLang-Omni from main, `sgl-omni serve --model-path inclusionAI/Ming-3.0-audio-tiny --talker.gpu 0`, which exposes OpenAI-style /v1/audio/transcriptions, /v1/audio/speech and /v1/chat/completions
- For transcripts use greedy decoding with 10-gram blocking and the validated ASR prompt, and expect a language label, a tab, then the text
- Parse tool calls yourself from `<tool_call>name` with `<arg_key>` and `<arg_value>` pairs, since no runtime parser is named
- Thinking is off by default, turn it on with chat_template_kwargs enable_thinking true and strip the `<think>` block from the reply yourself
- In vLLM-Omni pick a voice with audio.voice id from docs/voices.md, and don't combine it with a reference recording
- Review the repository's Python before passing trust_remote_code, and consider deleting the __pycache__ folders under talker/ so only readable source runs

## What the review panel asked for

- dated release notes
- a 4-bit build
- a tool-calling example
- a named parser
- remove the compiled .pyc files
- 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.