Granite Speech 5.0 470M TurboCTC is a IBM open-weight speech recognition (ASR) model. Real-time factor throughput averages 1×+ (Source not recorded (Aug 25, 2026)). CompareLLM speech recognition Elo sits at 1750. IBM Granite English speech recognition model with downloadable Apache-2.0 weights. Current transcription quality rating is an editorial estimate, not imported or reproduced word-error-rate evidence. Numbers below reflect verified disclosures and empirical benchmark harnesses.
Independent evaluation answering: "Is Granite Speech 5.0 470M TurboCTC the right model for your workload & budget?"
Specialized speech-to-text model; engineered for acoustic audio transcription, not multi-turn conversational chat or text reasoning.
Dated snapshot metrics aggregated from official evaluators and API providers with visual relative score bars.
Multiple of real-time audio processed. 35x RTFx transcribes a 1-hour audio stream in under 105 seconds.
Percentage of words inserted, deleted, or substituted. Lower WER indicates superior transcription accuracy.
Pairwise preference rating across noisy acoustics, overlapping speakers, and diverse accents.
Acoustic frame alignment architecture engineered for non-autoregressive fast streaming audio decoding.
Compact model sizing allows concurrent multi-channel transcription on consumer GPUs or on-device edge compute.
Local on-premise execution without transmitting proprietary or medical audio recordings to external cloud APIs.
| Benchmark Metric & Meaning | Reported Score & Capability Fill | CompareLLM Review |
|---|---|---|
Open ASR Leaderboard Real-Time Factor (English) Transcription Speed | 1×+ | Source not recorded · Aug 25, 2026Vendor-reported Open ASR public English test sets, 1 H200 GPU, results as of 2026-08-25. Model-card chart average RTFx 13042.98. Not an on-device latency claim or CompareLLM measurement. |
Select any rival to launch a side-by-side empirical benchmark comparison with winner deltas.
Compare Granite Speech 5.0 470M TurboCTC against
Models may use different benchmarks and test settings. This is an indicative composite, not a controlled head-to-head comparison or community Elo. Admin-approved sentiment estimates fill categories without accepted benchmark results. Estimates are labelled and do not increase benchmark coverage. Coverage refers to the configured recipe, not confidence.
Release recent-models-2026-09-23-r1 · recipe reported-text-2026-09-11-r1 · method reported-with-estimates-v2 · research through 2026-09-23
Overall unrated: insufficient applicable categories or no cross-task Overall.
Proposed editorial score for English speech recognition only: vendor evaluation claims and early practitioner interest indicate useful transcription quality, but benchmark numbers have not been imported or independently reproduced here. No multilingual competence is implied.
As of 2026-09-12 · review on 2026-09-26 · Editorial estimate (not a benchmark result)
Sentiment source 1 →Sentiment source 2 →What this model is available as, what it takes to run, and where its identity comes from.
The weights are published for download, so this model can be run on your own hardware.
Published weights — the provider's own download page.
Concise empirical overview formatted for citations and prompt context
Plain-English methodology and leaderboard answers
Follow this model in your watchlist, set it as your global comparison baseline, or assign it to your custom production stack.
Track updates & rank changes
Compare all models against this
Assign to custom architecture
Compare side-by-side vs all
Plotted against all active catalog models (50th percentile = catalog median).
1 of 3 dimensions have ranking-eligible evidence. Missing axes are not converted to zero.
Ranks in the top tier (≥75th percentile) for Speech recognition.
