live wire
▸AI · Red Hat documents usage-based admission fair sharing for Kueue 1.4 on OpenShiftRed Hat Developer▸AI: Red Hat maps governed firewall changes from ServiceNow through Ansible and two human approval gatesRed Hat Developer▸CLUSTER MGMT · ACM 2.17 makes Submariner 0.24 GA with Important-rated fixesRed Hat Errata▸PLATFORM · Red Hat makes on-premises Lightspeed recommendations GA for Satellite 6.18Red Hat Errata▸SECURITY · Red Hat Hardened Images updates Tomcat 10 for nine authentication, access-control and DoS flawsRed Hat Errata▸AI · Open Data Hub 3.6.0 EA1 bundles Trainer, MLflow and llm-d componentsOpen Data Hub▸AI · Speculators 0.6.0 adds P-EAGLE parallel drafting for vLLM speculative decodingRed Hat Developer▸SECURITY · OpenShift 4.17.57 fixes seven Go and TLS CVEs in an Important-rated updateRed Hat Errata▸AI · Red Hat benchmarks local LLM guardrails with EvalHub, exposing regex accuracy and latency trade-offsRed Hat Developer▸AI · Red Hat maps silent tool-call failures across agentic pipelinesRed Hat▸API · Kuadrant 1.5.3 adds GRPCRoute policies and developer-portal API-key workflowsKuadrant▸AI · (Aug 25) IBM releases Apache-2.0 Granite 4.2 reasoning models in 3B, 8B and 30B sizesIBM Research▸JAVA · Red Hat build of Quarkus 3.33.3.SP1 fixes 13 CVEs in an Important-rated updateRed Hat errata▸AI · vLLM moves Kimi K2 RL weight sync across 384 H100s in 7.53 seconds (Aug 22)vLLM▸AI · Red Hat documents usage-based admission fair sharing for Kueue 1.4 on OpenShiftRed Hat Developer▸AI: Red Hat maps governed firewall changes from ServiceNow through Ansible and two human approval gatesRed Hat Developer▸CLUSTER MGMT · ACM 2.17 makes Submariner 0.24 GA with Important-rated fixesRed Hat Errata▸PLATFORM · Red Hat makes on-premises Lightspeed recommendations GA for Satellite 6.18Red Hat Errata▸SECURITY · Red Hat Hardened Images updates Tomcat 10 for nine authentication, access-control and DoS flawsRed Hat Errata▸AI · Open Data Hub 3.6.0 EA1 bundles Trainer, MLflow and llm-d componentsOpen Data Hub▸AI · Speculators 0.6.0 adds P-EAGLE parallel drafting for vLLM speculative decodingRed Hat Developer▸SECURITY · OpenShift 4.17.57 fixes seven Go and TLS CVEs in an Important-rated updateRed Hat Errata▸AI · Red Hat benchmarks local LLM guardrails with EvalHub, exposing regex accuracy and latency trade-offsRed Hat Developer▸AI · Red Hat maps silent tool-call failures across agentic pipelinesRed Hat▸API · Kuadrant 1.5.3 adds GRPCRoute policies and developer-portal API-key workflowsKuadrant▸AI · (Aug 25) IBM releases Apache-2.0 Granite 4.2 reasoning models in 3B, 8B and 30B sizesIBM Research▸JAVA · Red Hat build of Quarkus 3.33.3.SP1 fixes 13 CVEs in an Important-rated updateRed Hat errata▸AI · vLLM moves Kimi K2 RL weight sync across 384 H100s in 7.53 seconds (Aug 22)vLLM
upstreambeat.ai
releaseAI

IBM’s Granite Speech 5.0 targets high-throughput English transcription

Two 470M-parameter encoder-only models trade licensing and a small accuracy difference for unusually high vendor-reported throughput.

IBM speech models with throughput, error rate, and license differences.
Chart: figures from the story
By The News Desk· Aug 25, 2026

IBM has released two Granite Speech 5.0 Turbo CTC models for English automatic speech recognition. Both have 470 million parameters, while one is licensed under Apache 2.0 and the other uses CC-BY-NC-SA-4.0 because it was trained on additional data (IBM Granite).

The release is aimed at high-volume transcription and edge-oriented speech-to-text rather than the broader speech capabilities of IBM's earlier language-model-backed Granite Speech systems. IBM reports throughput above 12,600 times real time on an Nvidia H200 with batched inference—more than 3.5 hours of audio transcribed in one second—but labels its public OpenASR measurements unofficial pending leaderboard results (IBM Granite).

Speed, accuracy and the license split

On the public short-form test sets used by the OpenASR Leaderboard, IBM reports an aggregate word error rate of 5.00% for the Apache 2.0 model and 4.85% for the noncommercial variant. The company says both exceeded 12,600 RTFx in its measurements. Those numbers were produced with Hugging Face Jobs and OpenASR tooling, but they remain vendor-reported results rather than an independent production evaluation (IBM Granite).

The licensing difference is operationally significant. IBM says the noncommercial model used additional GigaSpeech and SPGI Speech training data. The Apache 2.0 model used a smaller set of natural speech corpora, alongside synthetic data, and is the straightforward candidate when permissive commercial use is required (IBM Granite).

A narrower architecture

Unlike earlier Granite Speech models, the 5.0 Turbo CTC pair is encoder-only. IBM describes a stack of 16 Conformer blocks with self-conditioning, chunkwise attention and three stages of temporal subsampling. The design emits 12.5 tokens per second rather than the 50 characters per second used by prior encoders, which the company credits for much of the throughput gain (IBM Granite).

That narrower design has tradeoffs. IBM says the models give up speech translation and keyword biasing available in its language-model-equipped systems. It positions the new pair for transcription on edge devices, where a smaller memory footprint and fast inference can matter more than those broader features (IBM Granite).

IBM says native Hugging Face Transformers support is available from source until the next Transformers release. Before deployment, teams should validate accuracy on their own accents, recording conditions and domain vocabulary, and should treat the two model licenses as separate product-selection constraints rather than minor packaging differences.

sources

Filed by The News Desk. Corrections: desk@upstreambeat.ai · Our standards →

comments · 0

    Comments are moderated before they appear. Your email is used once to confirm it is you — never shown, never sold. Corrections and questions get an answer from the desk when we have one.

    more from the desk