whatmodel.sh / tts / Gepard 1.0
Gepard 1.0
↗ share
+ contribute
Gepard 1.0 is an open-weights text-to-speech model from nineninesix.ai, released June 2026. Its published size is 0.556B parameters. Licence: apache-2.0.
Catalogue identity notes
Comparison-only for now: exact checkpoint nineninesix/gepard-1.0 @ 440c3a2 (Apache-2.0, created 2026-06-22; stop_head weights updated 2026-08-06, other parameters unchanged per card). The card's and arXiv:2609.04222v1 §6.3 table row 'Gepard 1.0 (Ours)'. Distinct from gepard-1.1 (2026-09-05). The official inference stack (github.com/nineninesix-ai/gepard-inference) is a CUDA server + client ('make serve' in one terminal, 'make generate' in another), so no single copy-paste command was added; a hosted API exists at nineninesix.ai but no price was verified.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
Card: 'Parameters: ≈ 555.7M (backbone + audio interface + voice-cloning compressor)'; Qwen3.5-based 14-layer backbone ≈500M. NVIDIA NeMo NanoCodec (separate, NVIDIA Open Model License) not included.
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 Gepard 1.0, out of 148 with a score.
| # | model | score | where |
|---|---|---|---|
| 70 |  VibeVoice 1.5B | 53±6 | local |
| 71 |  Lightning v3.1 | 53±3 | api $17.50/M chars |
| 72 | Gepard 1.0 | 53±4 | comparison only |
| 73 |  Kokoro 82M v1.0 | 52±3 | api $0.62/M chars +local |
| 74 |  LEMAS-TTS (w/ prosody, multilingual_prosody) | 52±6 | local |
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
voice cloning▁·▁▁▁▁▁▃██▃▃ #34 of 43 · 94% of its score · 3 tests
Beats CosyVoice-300M here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Seed-TTS test-en · speaker similarity (WavLM-large SV) lab | 58.5% ✎ | ▁·▃▂▂█▆▃▅··▂ beats 2/30 | 40% |
| Seed-TTS test-en WER (transcriber not stated) lab | 3.6% ✎ | ▃··········█ beats 0/3 | 39% |
| Seed-TTS test-en · UTMOS (predicted MOS) lab | 2.64 / 5 ✎ | █·····██···█ beats 2/3 | 16% |
$ where to run it
No verified API or download recorded yet · add one
Official weights and documentation: huggingface.co/nineninesix ↗
$ how the score was worked out · nineninesix/gepard-1.0 model… 94%
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 |
|---|---|---|
 nineninesix/gepard-1.0 model card (Seed-TTS-eval benchmark table) · lab-reported | 3 | 94% |
| prior · pull toward the middle | 6% | |
| weighted average | = 52.5 |
$ 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