whatmodel.sh / tts / GLM-TTS
GLM-TTS
↗ share
+ contribute
GLM-TTS is an open-weights text-to-speech model from Z.ai, released December 2025. Its published size is 1.55B parameters. Licence: mit.
Catalogue identity notes
zai-org/GLM-TTS @ b3feb691fa4cf590eec4418f6275f95220e58844 (created 2025-12-10, MIT): the released base checkpoint (LLM + flow-matching + vocoder, zero-shot voice cloning, zh/en). The README lists 'Model Weights Optimized via Reinforcement Learning' as coming soon, so this is GLM-TTS (Base), not GLM-TTS_RL. Results: OpenBMB UltraEval-Audio replication/GLM-TTS.md (registry config glmtts, ckpt_dir zai-org/GLM-TTS, use_phoneme False). Code repo licence Apache-2.0; weights MIT per the HF card.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
Stage-1 LLM only: OUR COUNT from the safetensors headers of llm/model-00001-of-00002 and -00002 at the pinned revision (1,552,009,216 F32 parameters; LlamaForCausalLM, 28 layers, hidden 2048). The flow-matching model (flow/flow.pt), speech tokenizer and vocoder are not included; the card states no parameter count. Observed 2026-10-09.
Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.
$ how it compares
The text-to-speech models ranked either side of GLM-TTS, out of 148 with a score.
| # | model | score | where |
|---|---|---|---|
| 32 |  IndexTTS-2 | 63±1 | local |
| 33 |  MOSS-TTS v1.5 (8B) | 63±2 | local |
| 34 | GLM-TTS | 63±2 | local |
| 35 |  Sonic 3.5 | 63±2 | comparison only |
| 36 |  MiniMax Speech-02-HD | 63±3 | api $100.00/M chars |
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
voice cloning▁·▁▁▁▁▁▁▇█▃▃ #25 of 37 · 38% of its score · 4 tests
Trails F5-TTS Base (v0), E2 TTS Base (F5-TTS reproduction), DMOSpeech 2 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Seed-TTS test-en · speaker similarity (WavLM-large SV) | 67.3% ✎ | █·█·██·▄·▄·▄ beats 2/10 | 14% |
| Seed-TTS test-zh · speaker similarity (WavLM-large SV) | 76.0% ✎ | ▃··▃█▅▅··▃·▃ beats 2/10 | 14% |
| Seed-TTS test-en · Whisper large-v3 WER | 2.5% ✎ | ▃··▃▅··▅·█▃▅ beats 3/11 | 5.4% |
| Seed-TTS test-zh · Paraformer CER | 1.1% ✎ | ▂··▄··▂▃█·▄▂ beats 4/11 | 4.8% |
other tests5 not in a task · 60% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CV3-Eval hard-en · WER | 9.4% ✎ | █··██·██████ beats 3/8 | 18% |
| CV3-Eval ZH (zero-shot Chinese) · CER | 3.6% ✎ | █▄···█▄·█··▄ beats 4/8 | 15% |
| CV3-Eval hard-zh · CER | 9.1% ✎ | ▃▃·▃··▃▃█··▃ beats 3/8 | 12% |
| CV3-Eval hard-en · speaker similarity (ERes2Net SV) | 73.4% ✎ | ▄█▄·▄··▄··▄█ beats 6/8 | 11% |
| CV3-Eval EN (zero-shot English) · WER | 6.6% ✎ | ▃·······▃▅█▅ beats 3/8 | 4.7% |
$ where to run it
Other ways to run it
- Run it as published · needs ~8.89 GB, NVIDIA only · from github.com/zai-org
git clone https://github.com/zai-org/GLM-TTS.git && cd GLM-TTS && pip install -r requirements.txt && pip install -U huggingface_hub && huggingface-cli download zai-org/GLM-TTS --local-dir ckpt && python glmtts_inference.py --data=example_zh --exp_name=_test --use_cache
Weights and documentation: huggingface.co/zai-org ↗
$ how the score was worked out · UltraEval-Audio (OpenBMB) · T… 99%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 UltraEval-Audio (OpenBMB) · TTS replication results | 9 | 99% |
| prior · pull toward the middle | 1% | |
| weighted average | = 62.9 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json