Ming-3.0-tts-0.5B

# Fix list: Ming-3.0-tts-0.5B

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/ming-3-0-tts-0-5b, the October 2026 research run, assessed 21 October 2026. Grade C, 57.8 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-tts-0.5B: 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, 66 out of 100, up to 11.9 more on the total

Why it scored 66: From 2 independent benchmark results in 1 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, Intelligibility and speaker similarity 66. 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. Reach, 55 out of 100, up to 11.3 more on the total

Why it scored 55: Memory 30, the only build is inclusionAI's BF16 checkpoint at 2.99 GB with its speaker encoder, about 3.6 GB with the fifth added, far under 16 GB. Runtimes 0, none of the six named runtimes runs this architecture from its own code. Transformers has no BailingMMNative model and the config's auto_map points to a file that lives only in inclusionAI's GitHub repository, so it loads through the publisher's scripts, and vLLM-Omni is a separate runtime the line doesn't name. mlx-audio names the model but isn't on the list either. Builds 0, BF16 only, and every quantised copy is someone else's. Terms 15, plain MIT, since the card's disclaimer asks rather than binds. Download 10, the repository isn't gated. Hosted 0, no inference provider on its Hugging Face page and no public API documented by inclusionAI or Ant Group. Artificial Analysis shows an Ant Group host with no price, which couldn't be checked as a public API, so it's scored as absent and left as an open question.

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.

## 3. Risk, 55 out of 100, up to 9 more on the total

Why it scored 55: File safety 15, safetensors and ONNX with no pickle files, less 10 because the documented way to load it runs inclusionAI's own scripts, which build the prompt in code and pass trust_remote_code to the tokenizer. Provenance 15, inclusionAI's own organisation linked from its GitHub repository, with a SHA-256 for every LFS file and git hashes for the rest. Safety evaluation 0, none published. Agent guidance 5, the card's general cautions against impersonation and for labelling synthetic speech, with no safeguards described. Licence risk 15, MIT with nothing added and a working licence link. 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 disclosure policy, SECURITY.md or security.txt for inclusionAI 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.

## 4. Documentation and upkeep, 44 out of 100, up to 5.6 more on the total

Why it scored 44: Model card 14, it covers intended use, limitations, architecture, parameters and languages but gives no context length or knowledge cut-off. Training data 0, nothing described anywhere the publisher put it. Last change 20, the card and the inference code both changed on 19 October 2026. Responsiveness 0, the only Hugging Face discussion has no reply from inclusionAI, and GitHub issues couldn't be read, so this rests on one two-day-old thread. Version policy 5, a first release with the previous generation kept in its own repository and no dated note. File health 5, the tokenizer's template differs from the prompt the code builds and config.json's auto_map names a file that isn't in the Hub repository.

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, 55 out of 100, up to 4.5 more on the total

Why it scored 55: Read for a speech model, with the tool-calling lines taken as what a request declares and the structured-output line as what comes back. Chat template 10, the Jinja template in the repository is Qwen2.5's and unused, and the real prompt is built by inclusionAI's prompt_contract.py. Tool-calling format 25, the publisher documents every task, its fixed control values, the voice catalogue and the WAV output in its own Python API, CLI and cookbook. Trained tool calling 10, one task per call with no combining of controls across tasks. Structured output 5, a whole 44.1 kHz waveform with no streaming, timestamps or typed metadata. Reasoning control 0, none to control and no not-applicable rule. Context 5, the backbone's 32,768 positions in config.json, since the card states none.

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 enabled for this session, so responsiveness rests on one Hugging Face discussion with no reply and is scored 0
- Unchecked, whether Ant Group serves Ming-3.0-tts through a public API. Artificial Analysis shows an Ant Group host with no price and links only antgroup.com, so hosted is scored 0
- Whether a speech model's runtime line should credit vLLM-Omni or mlx-audio, which both name Ming TTS but aren't among the six runtimes the line names. Each would add 5 if the founder counts it
- The checklist has no not-applicable rule for reasoning control and knowledge cut-off on a speech model, so both cost points here as for the other Ming listings
- The Hugging Face repository carries no LICENSE or NOTICE of its own, and the card doesn't mention the licences of the bundled Qwen2.5-0.5B backbone or CAM++ speaker encoder
- Confidence follows Capability's independent areas, which the script computes. Artificial Analysis's arena Elo and speed are the only independent results found, and they aren't Seed-TTS-eval, so low is a placeholder
- 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 standard runtime runs it, only inclusionAI's GitHub scripts (mlx-audio and vLLM-Omni aside)
- Loading runs the publisher's code with trust_remote_code, and the prompt is built in code rather than a template
- One task per call, and a whole waveform back with no streaming or timestamps
- No safety evaluation, training data, quantised build or security route of its own
- No hosted route for an agent without a GPU

## 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.

- Clone github.com/inclusionAI/Ming-3.0-tts and `pip install -r requirements.txt` (torch 2.6.0, transformers 4.46.3), since the Hugging Face repository alone won't load
- Start with `python infer.py --model_path inclusionAI/Ming-3.0-tts-0.5B --task IP --text "The garden will remain open until sunset." --ip US_James --use_zero_spk_emb --no_text_normalization --output output/tts.wav`
- Pass `--no_text_normalization` unless pynini 2.1.6 is installed, since text normalisation imports it
- Pick one task per call and pass only that task's controls, since controls from another task are rejected
- Run `python tools/preflight.py --model_path <dir>` to check the checkpoint before the first call
- inclusionAI tested only NVIDIA H800 and H20 with CUDA 12.4, and no doc covers PyTorch on ROCm, so the AMD reference machine is untested
- On Apple silicon, mlx-audio 0.6.1 or later loads mlx-community's conversion, which isn't inclusionAI's
- Clone only voices you have consent for and label the audio as synthetic, as the card asks

## What the review panel asked for

- upload the auto_map file
- publish the technical report
- a stated memory figure
- streamed audio output
- watermarking for synthetic speech
- 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.