Public story · 2026-08-26 · high

IBM's speech model claims 3.5 hours of audio a second

Granite Speech 5.0's 470M model reaches 5% word error rate on English benchmarks, per IBM's unofficial numbers.

Why now: IBM published the model card on Hugging Face as of August 26.

Issue

2026-08-26

Confidence

high

Sources

Redaction

passed

Story

IBM released granite-speech-5.0-470m-turboctc, a 470 million parameter speech recognition model under Apache 2.0. The company reports over 12,600 RTFx on a single H200 GPU, roughly 3.5 hours of audio transcribed per second of compute.

That throughput number matters more than the model size. Most teams paying a cloud vendor per minute of audio are paying for compute they could run themselves, if IBM's 5.00% word error rate on OpenASR English short-form sets holds up outside its own testing.

The architecture stacks 16 conformer blocks trained with CTC under a 16,384 BPE token head. It subsamples the audio timeline by a factor of 8 and uses 128-frame block attention. A self-conditioned CTC pass off the middle layer feeds back into later layers. None of that is new by itself. IBM packed it into 470 million parameters and still reached that 5.00% word error rate.

IBM's own model card labels the numbers unofficial, pending an update to the public leaderboard. That's a real caveat. Vendor-reported throughput and accuracy figures often drift once independent runs replicate them on different hardware or batch sizes.

Source trail

granite-speech-5.0-470m-turboctc More from huggingface.co

Entities

Apache ASR BPE CTC H200 IBM OpenASR English RTFx Vendor WER

Claim evidence

IBM's speech model claims 3.5 hours of audio a second
granite-speech-5.0-470m-turboctc

Provenance

Canonical issue: 2026-08-26
AI generated: yes
Story unit: 2026-08-26-ibm-released-a-470m-apache-2-0-asr-model-claiming-12-600-rtfx-on-one-h200
Labels: source-backed, canonical briefing excerpt

Fetching from the wire…

IBM's speech model claims 3.5 hours of audio a second

Story

Related stories

Meta ships Muse Glimmer, a 30B open agent model

Agent Plugins 1.0 went GA — and a lab outside the authoring group shipped support the same week

The Linux Foundation launched Akrites to defend open source against AI-enabled threats.

Mistral ships a safety model that takes plain-English rules

Jensen Huang's First Tweet Doubles AI Coalition to 50

Qwen jumped from 1% to 69% of open-model fine-tunes

IBM will retrain tens of thousands of consultants on OpenAI's stack, less than a year after its Anthropic alliance.

NVIDIA forms security alliance, leaves out OpenAI and Anthropic

Source trail

Entities

Claim evidence

Provenance

IBM's speech model claims 3.5 hours of audio a second

Story

Related stories

Meta ships Muse Glimmer, a 30B open agent model

Agent Plugins 1.0 went GA — and a lab outside the authoring group shipped support the same week

The Linux Foundation launched Akrites to defend open source against AI-enabled threats.

Mistral ships a safety model that takes plain-English rules

Jensen Huang's First Tweet Doubles AI Coalition to 50

Qwen jumped from 1% to 69% of open-model fine-tunes

IBM will retrain tens of thousands of consultants on OpenAI's stack, less than a year after its Anthropic alliance.

NVIDIA forms security alliance, leaves out OpenAI and Anthropic

Source trail

Entities

Claim evidence

Provenance