Ming-3.0-audio-flash
- In Input modalities: TextAudio
- Open weights
Prices
0 offerings · prices as listedNo priced offering is listed for this model yet.
Details
- Family
- ming-audio-flash
Not stated on Ant Group’s pages: max input, output modalities, reasoning, reasoning efforts, reasoning can be off, max thinking budget, tool use, structured output, built-in tools, knowledge cutoff, licence, aliases.
Run locally
Weights: inclusionAI/Ming-3.0-audio-flash- Parameters 128.2B
- Active 6.5B · 512 experts, 8 per token
- Native precision BF16
- Layers 42
- KV cache per token 47 KB
- Config positions 128K (config.json max_position_embeddings)
- Needs
- ≈ 76.5 GB (weights 72.5 + KV cache 0.4 + overhead 3.6)
- Usable
- 21.6 GB (90% of its memory)
- In system memory
- 55.3 GB · 10/42 layers on the GPU (--n-gpu-layers 10)
- Generation speed
- 14–22 tok/s (reads the active 6.5B per token)
- First token
- about 0.21–0.31 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 238.8 GB | ≈ 251.1 | ✕Doesn’t fit | — |
| Q8_0 | 126.8 GB | ≈ 133.6 | ✕Doesn’t fit | — |
| FP8 | 119.4 GB | ≈ 125.7 | ✕Doesn’t fit | — |
| Q6_K | 98.0 GB | ≈ 103.3 | ✕Doesn’t fit | — |
| Q5_K_M | 84.9 GB | ≈ 89.5 | ✕Doesn’t fit | — |
| Q5_0 | 82.1 GB | ≈ 86.5 | ✕Doesn’t fit | — |
| Q4_K_M | 72.5 GB | ≈ 76.5 | ○Needs system memory 14–22 tok/s | — |
| Q4_0 | 67.2 GB | ≈ 70.9 | ○Needs system memory 16–24 tok/s | — |
| IQ4_XS | 63.4 GB | ≈ 67.0 | ○Needs system memory 17–26 tok/s | — |
| Q3_K_Mquality drops | 58.3 GB | ≈ 61.6 | ○Needs system memory 39–59 tok/s | — |
| IQ3_XXSquality drops | 45.7 GB | ≈ 48.3 | ○Needs system memory 50–75 tok/s | — |
| Q2_Kquality drops | 44.8 GB | ≈ 47.4 | ○Needs system memory 51–77 tok/s | — |
| IQ2_XXSquality drops | 30.7 GB | ≈ 32.6 | ○Needs system memory 74–112 tok/s | — |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): GPU NVIDIA RTX PRO 6000 Blackwell Workstation Edition; unified memory Apple M5 Max (40-core GPU) · 128 GB.
Run commands
The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated
Cited scores
From third-party leaderboards, not measured by usThis model isn’t on any leaderboard we cite yet.
News
All news →- Oct 19, 2026Release Ant Group releases three Ming audio models Ant Group launched Ming-3.0-asr-tiny, Ming-3.0-tts and Ming-3.0-audio-flash, available through its open API.
Changes
All changes →- New language model
Sources
- Changelog (opens in a new tab) developer.ant-ling.com · checked Nov 1, 2026
- Multimodal Understanding (opens in a new tab) developer.ant-ling.com · checked Nov 3, 2026
- inclusionAI/Ming-3.0-audio-flash config.json (opens in a new tab) huggingface.co · checked Nov 4, 2026
- inclusionAI/Ming-3.0-audio-flash model card (opens in a new tab) huggingface.co · checked Nov 4, 2026
- inclusionAI/Ming-3.0-audio-flash files (opens in a new tab) huggingface.co · checked Nov 5, 2026
- inclusionAI/Ming-3.0-audio-flash README.md (opens in a new tab) huggingface.co · checked Nov 5, 2026
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
Ant Group · China · developer.ant-ling.com
Related models
Snapshots, the same family and other models from this provider