Ming-3.0-tts-tiny

# Fix list: Ming-3.0-tts-tiny

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/ming-3-0-tts-tiny, the October 2026 research run, assessed 21 October 2026. Grade D, 52.2 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 Ming-3.0-tts-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, 50 out of 100, up to 19.2 more on the total

Why it scored 50: Memory 25, inclusionAI publishes only BF16 and Ollama has nothing, so the 18.0 GB of publisher files plus a fifth come to 21.6 GB, inside the 24 GB band. Runtimes 0, none of the six named runtimes runs the lingv3_tts architecture from its own code. Transformers only loads inclusionAI's remote code, and SGLang-Omni and vLLM-Omni are separate projects from SGLang and vLLM, with no Ming-3.0 code on vLLM-Omni's main branch anyway. Builds 0, the MLX quantisations are mlx-community's. Terms 15, plain MIT. Download 10, the repository isn't gated. Hosted 0, no inference provider is listed on its Hugging Face page and no publisher API serving it was 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, 55 out of 100, up to 13.8 more on the total

Why it scored 55: File safety 15, safetensors plus an ONNX speaker encoder and no pickle files, less 10 because loading runs the repository's Python model code and prompts are built by prompt_contract.py. Provenance 15, the inclusionAI organisation that inclusion-ai.org links to, with a SHA-256 for every large file. Safety evaluation 0, none published. Agent guidance 5, the card's general cautions against impersonation and unlabelled synthetic speech. Licence risk 15, MIT with nothing added. 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 in the publisher's repositories, no security.txt on inclusion-ai.org and no 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, 44 out of 100, up to 8.6 more on the total

Why it scored 44: Model card 14, it covers intended use, limitations, architecture, parameters and languages but not input length or a knowledge cut-off. Training data 0, nothing described. Last change 20, the card changed on 20 October 2026. Responsiveness 0, the one question on Hugging Face has no reply from inclusionAI, and GitHub issues couldn't be read, so this is couldn't check scored as absent for that part. Version policy 5, a first release. File health 5, the card's temperature of 0 for generate() disagrees with the 1.5 in the config files, and the shipped Jinja template isn't the model's prompt format.

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 6.2 more on the total

Why it scored 60: Read for a speech model, so the tool-calling and structured-output lines ask what it takes and returns. Chat template 10, the prompt is built by the publisher's code, and the Jinja file in the repository is the backbone's unused template. Format 25, the card's task table lists the inputs for each of ten tasks and the SGLang-Omni section gives the request fields. One call 10, one task per call with no combining. Typed output 15, response_format with wav, mp3, flac or pcm is documented with how to ask, though there are no timestamps or streaming. Reasoning control 0, nothing documented and no not-applicable rule. Context 0, the backbone has 4,096 positions and the card states no length.

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: GitHub issues on inclusionAI/Ming-3.0-tts, since the GitHub API isn't reachable from this session; responsiveness rests on the one Hugging Face discussion
- 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
- unchecked: whether vLLM-Omni support exists on a branch or pull request, as the publisher's roadmap ticks it; scored 0 either way, since vLLM-Omni isn't vLLM
- Capability has no independent result for this model; the Artificial Analysis Ming-3.0-tts entry links to the 0.5B weights
- No sameModel, as no inclusionAI or Ant Group hosted API for it was found

## Weaknesses

- No runtime among llama.cpp, Ollama, vLLM, SGLang, LM Studio or Transformers runs it from its own code
- Loading runs the publisher's Python code, pinned to transformers 4.46.3, torch 2.7.1 and CUDA 11.8
- No quantised build from inclusionAI and no hosted provider
- No training data or safety evaluation, and no security route of its own
- One task per call, no streaming and no timestamps
- The card's generate() temperature disagrees with the config files

## 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.

- Install from the GitHub repository, `git clone https://github.com/inclusionAI/Ming-3.0-tts.git && cd Ming-3.0-tts && pip install -r requirements-tiny.txt`, in its own Python 3.10.15 environment, since the 0.5B model pins different versions
- Quick start is `bash examples/one_click/run_seed.sh`, which clones from the bundled reference clip
- To serve, run `sgl-omni serve --model-path inclusionAI/Ming-3.0-tts-tiny` from SGLang-Omni's main branch and POST to /v1/audio/speech with input, response_format and, for cloning, ref_audio plus ref_text
- Pick one task per call (seed, multilingual, paralinguistic, podcast, basic, emotion, dialect, freeform, voice_design, ip) and pass only the inputs that task accepts
- Expect one complete audio body, since there is no streaming, and no timestamps
- Pass temperature yourself, since the reference generate() uses 0 and SGLang-Omni takes 1.5 from the config
- Run `python tools/preflight_tiny.py --model_path <dir>` on the downloaded checkpoint before first use
- Label output as synthetic speech where it could pass for a real recording, as the card asks

## What the review panel asked for

- align card and config
- drop the unused template
- a 4-bit build
- state the maximum input length
- watermarking for synthetic speech
- a channel for abuse reports

## 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.