AssemblyAI-STT-3.7 vs Rumble-2-Flash
Compare documented capabilities and access. A dash means a comparable figure is unavailable.
| Specification | ||
|---|---|---|
| Task | Speech recognition | Speech recognition |
| Published capabilities | Recorded audio. Typically handles 32 languages, with optional fallback covering 99 languages. | Qwen3-ASR-based transcription for over 40 languages. Reson8 reports 6.6% English word error rate and 6.96% mean error on the multilingual Open ASR Leaderboard track. |
| Model size | Not published | 0.5B parameters |
| Access | Pre-recorded Speech-to-Text API as assemblyai-stt-3-7 | Public weights for Transformers and vLLM, plus the Reson8 API. |
| Result | Batch transcript with speaker labels | Transcript with detected or selected language |
| Operating limits | Not published | Supports explicit language, digit formatting, filler retention and verbatim-repeat controls. |
| Billing terms | $0.21 USD per hour of audio on AssemblyAI. | Not published |
| Sources | Publisher documentation · Additional documentation | Model weights · Additional documentation |
| Inputs | Audio | Audio |
| Weights | Weights unconfirmed | Open weights · Apache-2.0 |
| Released | Nov 18, 2026 | Nov 18, 2026 |