Tencent HunyuanSpecialist modelOpen weights
Hy-SimulMT-30B-A3B
Summary
Hy-SimulMT continuously translates spoken conversation between Chinese and English. Tencent reports scores for FLEURS, CoVoST2, ACL6060, and RealSI-Sentence, plus a long-form evaluation on RealSI. The publisher reports about three seconds of StreamLAAL. This measures streaming translation lag, not time to first audio.
The released model variant outputs translated text. Spoken output and voice cloning require an additional configured TTS service. The Interleave variant remains a configuration of the model. The chart qualitatively compares results with many systems but gives no exact numerical value.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 6 published benchmark results
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
FLEURS translation score (Show measurement, test conditions, and source)
- Source value
- 90.3
- Score
- 90.3
- Metric
- Publisher score (metric definition and scale not stated)
- Unit
- points
- Benchmark version
- Official model-card comparison table, 2026 release
- Category
- simultaneous-speech-translation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-SimulMT model card, short-form results table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Hy-SimulMT row; bidirectional Chinese-English simultaneous speech translation; publisher says higher scores are better. The table does not name the score formula or scoring scale.
- Context
- The publisher does not define the score formula or scale in the table; do not interpret these as BLEU or percentage values.
CoVoST2 translation score (Show measurement, test conditions, and source)
- Source value
- 75.5
- Score
- 75.5
- Metric
- Publisher score (metric definition and scale not stated)
- Unit
- points
- Benchmark version
- Official model-card comparison table, 2026 release
- Category
- simultaneous-speech-translation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-SimulMT model card, short-form results table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Hy-SimulMT row; bidirectional Chinese-English simultaneous speech translation; publisher says higher scores are better. The table does not name the score formula or scoring scale.
- Context
- The publisher does not define the score formula or scale in the table; do not interpret these as BLEU or percentage values.
ACL6060 translation score (Show measurement, test conditions, and source)
- Source value
- 81.2
- Score
- 81.2
- Metric
- Publisher score (metric definition and scale not stated)
- Unit
- points
- Benchmark version
- Official model-card comparison table, 2026 release
- Category
- simultaneous-speech-translation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-SimulMT model card, short-form results table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Hy-SimulMT row; bidirectional Chinese-English simultaneous speech translation; publisher says higher scores are better. The table does not name the score formula or scoring scale.
- Context
- The publisher does not define the score formula or scale in the table; do not interpret these as BLEU or percentage values.
RealSI-Sentence translation score (Show measurement, test conditions, and source)
- Source value
- 84.7
- Score
- 84.7
- Metric
- Publisher score (metric definition and scale not stated)
- Unit
- points
- Benchmark version
- Official model-card comparison table, 2026 release
- Category
- simultaneous-speech-translation
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-SimulMT model card, short-form results table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Hy-SimulMT row; bidirectional Chinese-English simultaneous speech translation; publisher says higher scores are better. The table does not name the score formula or scoring scale.
- Context
- The publisher does not define the score formula or scale in the table; do not interpret these as BLEU or percentage values.
RealSI long-form StreamLAAL (Show measurement, test conditions, and source)
- Source value
- 3
- Score
- 3
- Metric
- StreamLAAL streaming translation lag
- Unit
- seconds
- Benchmark version
- RealSI long-form audio evaluation, 2026 model-card release
- Category
- simultaneous-speech-translation-latency
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Tencent Hy-SimulMT model card, RealSI long-form evaluation chart and text
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Long-form Chinese-English translation on RealSI. The model card reports about 3 seconds; lower streaming lag is better.
- Context
- StreamLAAL measures streaming translation lag, not time to first speech output. The card does not publish a more precise value.
Matched interpreter study StreamLAAL (Show measurement, test conditions, and source)
- Source value
- 3.3
- Score
- 3.3
- Metric
- StreamLAAL streaming translation lag
- Unit
- seconds
- Benchmark version
- Matched human-interpreter comparison, 18 recordings
- Category
- simultaneous-speech-translation-latency
- Direction
- Lower is better
- Sample
- 18
- Source type
- vendor-reported
- Evaluator
- Tencent matched-interpreter chart embedded in the Hy-SimulMT model card
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Hy-SimulMT result in Tencent's comparison with six professional interpreters on 18 recordings of about six minutes each; the figure shows 3.3 seconds StreamLAAL. CATTI Level 1 and Level 2 values use the best score among three interpreters at each level.
- Context
- This small study applies to these 18 recordings, the participating interpreters, and the stated setup. StreamLAAL measures lag, not time to first speech.
All specifications
Architecture
Input and output
Licence and access
Further details
- Model output
- Translated text. Spoken output and voice cloning require a configured external TTS service.
- Source: Model output
- Interleave variant
- Returns source text and translation together; a configuration or variant, not a separate model profile
- Source: Interleave variant
- Speech rendering
- External TTS service required (Translated text is the model output; speech rendering and voice cloning use an external configured TTS service.)
- Source: Speech rendering
Sources and data date
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 29, 2026. 59 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | Exact GEMBA numeric score Note The RealSI chart shows points without exact numeric labels; prose provides only relative comparisons. |
| Not found in the inspected sources | Exact parameter counts Note The reviewed official model card and repository do not state total and active parameter counts separately. The model name alone is not treated as numeric evidence. |
| Not published | Short-form score formula Note The table says higher scores are better but does not name the metric or its scale. |
| Not found in the inspected sources | Vals rows in captured payloads Note Nine complete Vals payloads dated 28 October were checked; no exact model ID or unambiguous model label. Limited to these saved benchmarks. |
| Unresolved | Uncaptured Arena Code/Speech leaderboards Note Text, Agent including Code/Chat/Work, and twelve WebDev views were captured. A separate general Code and Arena Speech leaderboard were not saved as complete tables. |
| Unresolved | Artificial Analysis catalog reconciliation Note The full current AA model and benchmark catalog was not re-reconciled for this profile in this sidecar. This sidecar makes no AA absence claim; native AA imports remain authoritative. |
| Additional source | Tencent Hy-SimulMT model card and evaluation (retrieved October 29, 2026) · Tencent Hy-SimulMT model card and results · Editorial description reviewed October 29, 2026 |
| Additional source | Tencent Hy-SimulMT inference and training repository (retrieved October 29, 2026) · Tencent Hy-SimulMT inference and training repository · Editorial description reviewed October 29, 2026 |
| Additional source | Tencent matched interpreter StreamLAAL chart (retrieved October 29, 2026) · Tencent matched interpreter StreamLAAL chart · Editorial description reviewed October 29, 2026 |
| Additional source | Model metadata (retrieved October 29, 2026)12 |